

<!-- formatList: 

 Array
(
    [a* algorithm] =&gt; Array
        (
            [slug] =&gt; a* algorithm
            [parent] =&gt; 
            [name] =&gt; A* algorithm
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; 
        )

    [abduction] =&gt; Array
        (
            [slug] =&gt; abduction
            [parent] =&gt; 
            [name] =&gt; abduction
            [canonical] =&gt; abductive reasoning
            [see] =&gt; 
            [description] =&gt; 
        )

    [abductive reasoning] =&gt; Array
        (
            [slug] =&gt; abductive reasoning
            [parent] =&gt; 
            [name] =&gt; abductive reasoning
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; Abductve reasoning considers various explainations for observed situation and tries to find the simplest or most likely.  It is a {{knowledge-rich}} method of reasoning that can make ({{defeasible}}) deductions beyond the incontovertable facts (used by {{deductive reasong}}) and available empirical data (used by {{inducdtuve reasoning}}).  As sush it is a form of {{nin-monotonic reasoning}} as new knowledge could change the potetial explanations.  
        )

    [accountable] =&gt; Array
        (
            [slug] =&gt; accountable
            [parent] =&gt; 
            [name] =&gt; accountable
            [canonical] =&gt; accountability
            [see] =&gt; 
            [description] =&gt; 
        )

    [accountability] =&gt; Array
        (
            [slug] =&gt; accountability
            [parent] =&gt; 
            [name] =&gt; accountability
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; This refers to legal or other means by which companies, governments or individuals are held to account for AI algorithms or systems that they produce or deploy.  If something goes wrong they may be taken to court, fined, imprisoned or at least be subject to public approbation.  It is believed if that the costs of negligence are high enough, then companies and other actors will act responsibly.
        )

    [accuracy] =&gt; Array
        (
            [slug] =&gt; accuracy
            [parent] =&gt; 
            [name] =&gt; accuracy
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; Accuracy is the difference between the output or prediction of an algorithm and the true value.  For numerical results it is often measured using the {{mean square error}} (the average of the square of the difference) or absolute difference.  For classifications there are several different kinds of measures that are important, including {{precision}} and {{recall}} for binary decisions, so the word 'accuracy can be ambiguous.
        )

    [actstar] =&gt; Array
        (
            [slug] =&gt; actstar
            [parent] =&gt; 
            [name] =&gt; ACT&lt;sup&gt;*&lt;/sup&gt;
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; ACT&lt;sup&gt;*&lt;/sup&gt; is a cognitive architecture developed by {{John Robert Anderson}} and others in the mid-1970s.  It and its successor {{ACT-R}} have been highly influential within cognitive science..
        )

    [act*] =&gt; Array
        (
            [slug] =&gt; act*
            [parent] =&gt; 
            [name] =&gt; act*
            [canonical] =&gt; actstar
            [see] =&gt; 
            [description] =&gt; 
        )

    [act*!proceduralization] =&gt; Array
        (
            [slug] =&gt; act*!proceduralization
            [parent] =&gt; act*
            [name] =&gt; proceduralization
            [canonical] =&gt; act-r
            [see] =&gt; 
            [description] =&gt; 
        )

    [act-r] =&gt; Array
        (
            [slug] =&gt; act-r
            [parent] =&gt; 
            [name] =&gt; ACT-R
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; ACT-R (formally known as {{ACT*}}) has been developed by {{John Robert Anderson}} and others over the past 40 years. It comprises a large long- term memory in the form of a {{semantic net}}, a small working memory of active items and a production system that operates on the memories.   ACT-R includes mechanisms (i) to make commonly used instances of general rules more specific (called {{proceduralisation}}) and also to take multiple specifis rules with similar structure and create a single more abstract rule (called {{generalisation}}).
        )

    [act-r!generalization] =&gt; Array
        (
            [slug] =&gt; act-r!generalization
            [parent] =&gt; act-r
            [name] =&gt; generalization
            [canonical] =&gt; act-r
            [see] =&gt; 
            [description] =&gt; 
        )

    [active vision] =&gt; Array
        (
            [slug] =&gt; active vision
            [parent] =&gt; 
            [name] =&gt; active vision
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; Active vision is when the camera or camera platform changes its orientation or position in order to better recogonise or interpret and object or scene.  For example, you might look at an object from several angles, or even peek aorund the corner of a building.  In some cases a similar effect can be achieved by manipulating the object itself if it is held in a robotoc hand, or is on a moving platform.
        )

    [activity recognition] =&gt; Array
        (
            [slug] =&gt; activity recognition
            [parent] =&gt; 
            [name] =&gt; activity recognition
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; This refers to situations where an AI system needs to recognise some form of human activity, for example, cooking food in kitchen, or leavinga. Suspicious package at an airport.  Activity recognition systems often have a camera and/or some other form of sensors and have to both identifiy aspects of a single frame or reading, but also the changing pattern over time.
        )

    [ada lovelace] =&gt; Array
        (
            [slug] =&gt; ada lovelace
            [parent] =&gt; 
            [name] =&gt; Ada Lovelace
            [canonical] =&gt; 
            [see] =&gt; lovelace, ada
            [description] =&gt; 
        )

    [adjacency statistics] =&gt; Array
        (
            [slug] =&gt; adjacency statistics
            [parent] =&gt; 
            [name] =&gt; adjacency statistics
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; In a text or in general a sequence, adjacency statistics measure the frequency with which words or tokens occur next to each other.  For example, in the sequence BAAACBBAAC, the pair (or 2-gram) AA occurs three times, AC and  BA each occur twice, and BB and  CB each occur once.  No three tokens (3-gram) in the example occur more than once, but in longer texts 3-grams (such as the words 'artificial neural network') and longer sequences may recur.
        )

    [adversarial] =&gt; Array
        (
            [slug] =&gt; adversarial
            [parent] =&gt; 
            [name] =&gt; adversarial
            [canonical] =&gt; adversarial learning
            [see] =&gt; 
            [description] =&gt; 
        )

    [adversarial learning] =&gt; Array
        (
            [slug] =&gt; adversarial learning
            [parent] =&gt; 
            [name] =&gt; adversarial learning
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; Adversarial learning/training pitches two or more algorithms against each other in a competitive situation. Each adversary uses {{machine learning}} to attempt to gain advantage over the other.  It takes inspiration from {{zero-sum games}}, where one player's gaims are the other player's loss.   The technique haa been used in {{generative adversarial networks}} for image generation, for example in {{deep fakes}}, where one adversary tries to generate new images and the other to distinguish generated images from real ones.  {{Self play}} is a similar technique in {{game playing}} AI.
        )

    [adversarial techniques] =&gt; Array
        (
            [slug] =&gt; adversarial techniques
            [parent] =&gt; 
            [name] =&gt; adversarial techniques
            [canonical] =&gt; adversarial learning
            [see] =&gt; 
            [description] =&gt; 
        )

    [adversarial training] =&gt; Array
        (
            [slug] =&gt; adversarial training
            [parent] =&gt; 
            [name] =&gt; adversarial training
            [canonical] =&gt; adversarial learning
            [see] =&gt; 
            [description] =&gt; 
        )

    [agents] =&gt; Array
        (
            [slug] =&gt; agents
            [parent] =&gt; 
            [name] =&gt; agents
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; 
            [subentries] =&gt; Array
                (
                    [agents!email filtering] =&gt; Array
                        (
                            [slug] =&gt; agents!email filtering
                            [parent] =&gt; agents
                            [name] =&gt; email filtering
                            [canonical] =&gt; 
                            [see] =&gt; 
                            [description] =&gt; 
                        )

                    [agents!embodiment] =&gt; Array
                        (
                            [slug] =&gt; agents!embodiment
                            [parent] =&gt; agents
                            [name] =&gt; embodiment
                            [canonical] =&gt; 
                            [see] =&gt; 
                            [description] =&gt; 
                        )

                    [agents!intelligent filtering] =&gt; Array
                        (
                            [slug] =&gt; agents!intelligent filtering
                            [parent] =&gt; agents
                            [name] =&gt; intelligent filtering
                            [canonical] =&gt; 
                            [see] =&gt; 
                            [description] =&gt; 
                        )

                    [agents!machine learning] =&gt; Array
                        (
                            [slug] =&gt; agents!machine learning
                            [parent] =&gt; agents
                            [name] =&gt; machine learning
                            [canonical] =&gt; 
                            [see] =&gt; 
                            [description] =&gt; 
                        )

                    [agents!messages] =&gt; Array
                        (
                            [slug] =&gt; agents!messages
                            [parent] =&gt; agents
                            [name] =&gt; messages
                            [canonical] =&gt; 
                            [see] =&gt; 
                            [description] =&gt; 
                        )

                    [agents!methods] =&gt; Array
                        (
                            [slug] =&gt; agents!methods
                            [parent] =&gt; agents
                            [name] =&gt; methods
                            [canonical] =&gt; 
                            [see] =&gt; 
                            [description] =&gt; 
                        )

                )

        )

    [agents!email filtering] =&gt; Array
        (
            [slug] =&gt; agents!email filtering
            [parent] =&gt; agents
            [name] =&gt; email filtering
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; 
        )

    [agents!embodiment] =&gt; Array
        (
            [slug] =&gt; agents!embodiment
            [parent] =&gt; agents
            [name] =&gt; embodiment
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; 
        )

    [agents!intelligent filtering] =&gt; Array
        (
            [slug] =&gt; agents!intelligent filtering
            [parent] =&gt; agents
            [name] =&gt; intelligent filtering
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; 
        )

    [agents!machine learning] =&gt; Array
        (
            [slug] =&gt; agents!machine learning
            [parent] =&gt; agents
            [name] =&gt; machine learning
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; 
        )

    [agents!messages] =&gt; Array
        (
            [slug] =&gt; agents!messages
            [parent] =&gt; agents
            [name] =&gt; messages
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; 
        )

    [agents!methods] =&gt; Array
        (
            [slug] =&gt; agents!methods
            [parent] =&gt; agents
            [name] =&gt; methods
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; 
        )

    [agi] =&gt; Array
        (
            [slug] =&gt; agi
            [parent] =&gt; 
            [name] =&gt; AGI
            [canonical] =&gt; 
            [see] =&gt; artificial general intelligence
            [description] =&gt; 
        )

    [ai arms race] =&gt; Array
        (
            [slug] =&gt; ai arms race
            [parent] =&gt; 
            [name] =&gt; AI arms race
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; There is literally an arms race in the military use of AI, but also an AI arms race between commercial companies.   Having the best AI may give massive commercial advantage either in terms of user experience (such as the runaway dominance of Google search due to the {{PageRank}} algorithm) or for behind the scenes activity such as the sue in advertisment placement.
        )

    [ai winter] =&gt; Array
        (
            [slug] =&gt; ai winter
            [parent] =&gt; 
            [name] =&gt; AI winter
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; The history of AI is often seen in terms of relatve excitement and success followed by periods of disillusionment, called AI winters. The most recent, sometimes simply called 'the AI Winter' was between the mid 1990s, following the early promise and over-hyping of {{neural networks}}  and early 2010s when the devlopemnt of {{deep neural networks}} led to the current high interest in AI.  Some fear that over-inlfalted expections of what AI can achieve may lead to yet another AI winter, but the widespreadsuccessful practical applications of AI makes this seem less likely.
        )

    [alien intelligence] =&gt; Array
        (
            [slug] =&gt; alien intelligence
            [parent] =&gt; 
            [name] =&gt; alien intelligence
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; AI often appears 'intellogent', but in ways thata re very differet to the way a human would tackle the same problem.  One can think of this as 'alien intelligence', misquoting Start Trek, &quot;intelligence, but not as we know it&quot;.  Although light-hearted, this can be a useful way to consider AI in practical applications.
        )

    [alpha-beta pruning] =&gt; Array
        (
            [slug] =&gt; alpha-beta pruning
            [parent] =&gt; 
            [name] =&gt; alpha&amp;ndash;beta pruning
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; Alpha&amp;ndash;beta pruning is a vaiant of {{minimax search}} for {{zero-sum}} {{two player games}}.  It adopts a similar insight to {{branch and bound}} for simple {{tree search}}.  As there are two players, it keeps a 'best so far' value for each player.  By convention alpha is the best for the first player and beta for the second, but these swop as one moves down the {{game tree}} alernating the players' turns.   Imagine the frist player has exmaind a numer of possible moves and has a best so far move with score 10.  It starts to explore a new move M and looks at the second player's options.  As soon as it encounters one with the second player's score greater than -10 (so first player worse than 10), it can prune the whole of the branch associated with M as it is clearly less good thath best so far move.  A similar strategy is used {{recursively}} but alternating 'best so far' roles.   Note, like {{minimax}}, alpha&amp;ndash;beta pruning assumes a {{perfect opponent}}.
        )

    [alphafold] =&gt; Array
        (
            [slug] =&gt; alphafold
            [parent] =&gt; 
            [name] =&gt; AlphaFold
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; AlphaFold and its successor AlphaFold2 applied some of the technology originally designed for {{AlphaGo}} and applied this to protein folding.  While some of the behaviour of proteins is given by their chemical and molecular structure, the physical shape of a protein, the way it naturally folds given the order of molecular units, is also critical.  For example, this folded shape deterimines which parts of the molecular structure are exposed and which hidden inside and thus the extent to which they are free to interact with other molecules and engage in biological processs.  Determing the folded shape from the chemcal structure is highly complex, and often relied on experimentally observing the structure using electron microscopy and other methods.  AlphaFold revolutionised this field allowing new molecules to be tested for efficacy purely by computer simulation (in silco).  This means only a relatively small number need progress to full lab-based synthesis and testing.
        )

    [alphafold2] =&gt; Array
        (
            [slug] =&gt; alphafold2
            [parent] =&gt; 
            [name] =&gt; AlphaFold2
            [canonical] =&gt; alphafold
            [see] =&gt; 
            [description] =&gt; 
        )

    [alphago] =&gt; Array
        (
            [slug] =&gt; alphago
            [parent] =&gt; 
            [name] =&gt; AlphaGo
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; AlphaGo was one of the key moments of modern AI, when a computer Go player defeated the Go Grandmaster Lee Se-dol.  It had previosuly been believed that the complexity of Go, in particular the large {{branching factor}} of its {{search space}} made it intractable with even near-future technology.  AlphaGo was initially taught using records of large numbers of human games, but then adopted adversarial techniques to improve its performance.  AlphaGo uses traditional computer game techniques, such as {{Monte Carlo tree search}}, guided by heuristics from {{deep neural networks}}.  {{AlphaZero}}, the successor of AlphaGo, dispenses with the need for human training.
        )

    [alphazero] =&gt; Array
        (
            [slug] =&gt; alphazero
            [parent] =&gt; 
            [name] =&gt; AlphaZero
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; AlphaZero is the successor of {{AlphaGo}}, which defeated a Go grandmaster for the first time in 2016.    Whilst {{AlphaGo}} was initialy trained on human games, AlphaZero started from scratch with only a description of the rules of Go and thereafter learning entirely using {{adversarial training}}.
        )

    [ambiguity] =&gt; Array
        (
            [slug] =&gt; ambiguity
            [parent] =&gt; 
            [name] =&gt; ambiguity
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; 
            [subentries] =&gt; Array
                (
                    [ambiguity!in computer vision] =&gt; Array
                        (
                            [slug] =&gt; ambiguity!in computer vision
                            [parent] =&gt; ambiguity
                            [name] =&gt; in computer vision
                            [canonical] =&gt; 
                            [see] =&gt; ambiguous image
                            [description] =&gt; 
                        )

                    [ambiguity!lexical] =&gt; Array
                        (
                            [slug] =&gt; ambiguity!lexical
                            [parent] =&gt; ambiguity
                            [name] =&gt; lexical
                            [canonical] =&gt; 
                            [see] =&gt; 
                            [description] =&gt; {{Ambiguity}} at a lexical level in {{NLP}} is about words, for example &quot;Look at the bow&quot;&quot; might mean look at the front of a ship, at a weapon to shoot arrows or at a person bending before royalty.
                        )

                    [ambiguity!pragmatic] =&gt; Array
                        (
                            [slug] =&gt; ambiguity!pragmatic
                            [parent] =&gt; ambiguity
                            [name] =&gt; pragmatic
                            [canonical] =&gt; 
                            [see] =&gt; 
                            [description] =&gt; {{Ambiguity}} at a pragmatic level in {{NLP}} is about the meaning of an utterance in context.  For exmaple &quot;it is cold&quot; might be a statement about the temeperature or a request to close the door.
                        )

                    [ambiguity!referential] =&gt; Array
                        (
                            [slug] =&gt; ambiguity!referential
                            [parent] =&gt; ambiguity
                            [name] =&gt; referential
                            [canonical] =&gt; 
                            [see] =&gt; 
                            [description] =&gt; {{Ambiguity}} at a referential level in {{NLP}} is about waht tjinsg refer to, especially, pronouns.  For example, in &quot;The dog chased the cat. It was fast.&quot; is it the dog or the cat that is fast?
                        )

                    [ambiguity!semantic] =&gt; Array
                        (
                            [slug] =&gt; ambiguity!semantic
                            [parent] =&gt; ambiguity
                            [name] =&gt; semantic
                            [canonical] =&gt; 
                            [see] =&gt; 
                            [description] =&gt; {{Ambiguity}} at a semantic level in {{NLP}} is about the meaning of text, for example, &quot;I saw her duck&quot; might I saw her drop down suddenly, or that I saw her pet poultry.  Note here how the zsemantic ambiguity related to {{lexical ambiguity}} of 'her' (as accusative or possessive pronoun) and 'duck' (avoidance action or bird).
                        )

                    [ambiguity!syntactic] =&gt; Array
                        (
                            [slug] =&gt; ambiguity!syntactic
                            [parent] =&gt; ambiguity
                            [name] =&gt; syntactic
                            [canonical] =&gt; 
                            [see] =&gt; 
                            [description] =&gt; {{Ambiguity}} at a syntactic level in {{NLP}} is about structure and {{grammar}}.  For example, does &quot;the man with the dog that is angry&quot; refer to the owner of an angry dog or and angry owner of a dog?
                        )

                )

        )

    [ambiguity!in computer vision] =&gt; Array
        (
            [slug] =&gt; ambiguity!in computer vision
            [parent] =&gt; ambiguity
            [name] =&gt; in computer vision
            [canonical] =&gt; 
            [see] =&gt; ambiguous image
            [description] =&gt; 
        )

    [ambiguity!lexical] =&gt; Array
        (
            [slug] =&gt; ambiguity!lexical
            [parent] =&gt; ambiguity
            [name] =&gt; lexical
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; {{Ambiguity}} at a lexical level in {{NLP}} is about words, for example &quot;Look at the bow&quot;&quot; might mean look at the front of a ship, at a weapon to shoot arrows or at a person bending before royalty.
        )

    [ambiguity!pragmatic] =&gt; Array
        (
            [slug] =&gt; ambiguity!pragmatic
            [parent] =&gt; ambiguity
            [name] =&gt; pragmatic
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; {{Ambiguity}} at a pragmatic level in {{NLP}} is about the meaning of an utterance in context.  For exmaple &quot;it is cold&quot; might be a statement about the temeperature or a request to close the door.
        )

    [ambiguity!referential] =&gt; Array
        (
            [slug] =&gt; ambiguity!referential
            [parent] =&gt; ambiguity
            [name] =&gt; referential
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; {{Ambiguity}} at a referential level in {{NLP}} is about waht tjinsg refer to, especially, pronouns.  For example, in &quot;The dog chased the cat. It was fast.&quot; is it the dog or the cat that is fast?
        )

    [ambiguity!semantic] =&gt; Array
        (
            [slug] =&gt; ambiguity!semantic
            [parent] =&gt; ambiguity
            [name] =&gt; semantic
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; {{Ambiguity}} at a semantic level in {{NLP}} is about the meaning of text, for example, &quot;I saw her duck&quot; might I saw her drop down suddenly, or that I saw her pet poultry.  Note here how the zsemantic ambiguity related to {{lexical ambiguity}} of 'her' (as accusative or possessive pronoun) and 'duck' (avoidance action or bird).
        )

    [ambiguity!syntactic] =&gt; Array
        (
            [slug] =&gt; ambiguity!syntactic
            [parent] =&gt; ambiguity
            [name] =&gt; syntactic
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; {{Ambiguity}} at a syntactic level in {{NLP}} is about structure and {{grammar}}.  For example, does &quot;the man with the dog that is angry&quot; refer to the owner of an angry dog or and angry owner of a dog?
        )

    [ambiguous image] =&gt; Array
        (
            [slug] =&gt; ambiguous image
            [parent] =&gt; 
            [name] =&gt; ambiguous image
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; An ambiguous image is an image that has more than one interpretation.  A classic example is {{Rubin's vase}} which can be seen as a vase or two faces.   Human vision typically flips back and forth, largely unconstrollably, between the interpretations.  In real situatiosn the ambiguity eiether resolves naturally or one moves ones head to geta. different view.  The latter is emulated in {{actve visions}} where the camera moves.
        )

    [analogy] =&gt; Array
        (
            [slug] =&gt; analogy
            [parent] =&gt; 
            [name] =&gt; analogy
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; An analogy is when concept, item or situation is compared with another drawing out parallels.  For example, a school teacher with their pupils are like a shepherd herding sheep; in this (implicitluy) the teacher  is in the position of the shepherd and pupils rleated to the sheep.  There are usually limits beyind which the analogy does not hold, for example, one would not expect school childrden to grow wool.  Analogies are used as part of {{human reasoning}} to suggest potential solutions; and forms of this process can be adopted in AI systems  during {{reasoning by analogy}}.   {{Case-based reasoning}} systems may use {{reasoning by analogy}} to transform past cases.
            [subentries] =&gt; Array
                (
                    [analogy!derivational] =&gt; Array
                        (
                            [slug] =&gt; analogy!derivational
                            [parent] =&gt; analogy
                            [name] =&gt; derivational
                            [canonical] =&gt; 
                            [see] =&gt; derivational analogy
                            [description] =&gt; 
                        )

                    [analogy!transformational] =&gt; Array
                        (
                            [slug] =&gt; analogy!transformational
                            [parent] =&gt; analogy
                            [name] =&gt; transformational
                            [canonical] =&gt; 
                            [see] =&gt; transformational analogy
                            [description] =&gt; 
                        )

                )

        )

    [analogy!derivational] =&gt; Array
        (
            [slug] =&gt; analogy!derivational
            [parent] =&gt; analogy
            [name] =&gt; derivational
            [canonical] =&gt; 
            [see] =&gt; derivational analogy
            [description] =&gt; 
        )

    [analogy!transformational] =&gt; Array
        (
            [slug] =&gt; analogy!transformational
            [parent] =&gt; analogy
            [name] =&gt; transformational
            [canonical] =&gt; 
            [see] =&gt; transformational analogy
            [description] =&gt; 
        )

    [analytic rules] =&gt; Array
        (
            [slug] =&gt; analytic rules
            [parent] =&gt; 
            [name] =&gt; analytic rules
            [canonical] =&gt; 
            [see] =&gt; procedural knowledge
            [description] =&gt; 
        )

    [analytical engine] =&gt; Array
        (
            [slug] =&gt; analytical engine
            [parent] =&gt; 
            [name] =&gt; Analytical Engine
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; The Analytical Engine was a design by {{Charles Babbage}} in the 1830s following his {{Difference Engine}}, inspired by the punch-card-based  {{Jacquard Looms}}    It was never completed, but in concept is seen as the first programable computer.  {{Ada Lovelace}}, long term collaborator with Babbage idetified the special qualities of the ``{\em The Difference Engine can merely tabulate, and is incapable of developing, the Analytical Engine can either tabulate or develope.}''
        )

    [anderson, j] =&gt; Array
        (
            [slug] =&gt; anderson, j
            [parent] =&gt; 
            [name] =&gt; anderson, j
            [canonical] =&gt; anderson, john robert
            [see] =&gt; 
            [description] =&gt; 
        )

    [anderson, john robert] =&gt; Array
        (
            [slug] =&gt; anderson, john robert
            [parent] =&gt; 
            [name] =&gt; Anderson, John Robert
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; Anderson developed the highly influential {{ACT*}} cognitive architecture.
        )

    [andrey andreyevich markov] =&gt; Array
        (
            [slug] =&gt; andrey andreyevich markov
            [parent] =&gt; 
            [name] =&gt; Andrey Andreyevich Markov
            [canonical] =&gt; 
            [see] =&gt; markov, andrey
            [description] =&gt; 
        )

    [anonymisation] =&gt; Array
        (
            [slug] =&gt; anonymisation
            [parent] =&gt; 
            [name] =&gt; anonymisation
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; General Data Protection Regulation (GDPR) is the European Union {{data protection law}} intended, amongst other things, to protect the {{privacy}} of EU citizens.  It is regarded as one of the strongest {{data protection laws}} in the world and has international significane.  This is not just because the EU itself is a large market, but because it applied extra-terrotorially if internet or web-based systems are available for use in the EU.  This is of particular relevance in AI settings as it can be hard to be sure what information is preserved in the weights of a {{deep neural networks}}, emphasisng the importance of {{explainable AI}}. 
        )

    [anonymised] =&gt; Array
        (
            [slug] =&gt; anonymised
            [parent] =&gt; 
            [name] =&gt; anonymised
            [canonical] =&gt; anonymisation
            [see] =&gt; 
            [description] =&gt; 
        )

    [anonymous id] =&gt; Array
        (
            [slug] =&gt; anonymous id
            [parent] =&gt; 
            [name] =&gt; anonymous id
            [canonical] =&gt; anonymous identifier
            [see] =&gt; 
            [description] =&gt; 
        )

    [anonymous identifier] =&gt; Array
        (
            [slug] =&gt; anonymous identifier
            [parent] =&gt; 
            [name] =&gt; anonymous identifier
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; An anonymous identifier is a number or string that is unique to the person, but from which it is hard or impossibel to determine which person it refers to.  It is used, in {{pseudonymised}} data to help connect together different record that refer to the same indivdiual.  Sometiems, the lost that connects individuals with identifiers is deleted after use in order to prevent accidental disclosure, but sometime sit has to be retaianied in order to correctly pseudonymise further data.  In these case the {{authority file}} linking ids to people needs to be maintained at a much higher level of security to the {{pseudonomised data}} itself.
        )

    [anti-discrimination laws] =&gt; Array
        (
            [slug] =&gt; anti-discrimination laws
            [parent] =&gt; 
            [name] =&gt; anti-discrimination laws
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; Many nations have laws to prohibit particular forms of descriminiation often including gender, ethnicity and sexual orientation.  For example, the UK list of {{protected characteristics}} at the time of writing includes  age, gender reassignment, being married or in a civil partnership, being pregnant or on maternity leave, disability, race including colour, nationality, ethnic or national origin, religion or belief, sex, and sexual orientation.  The laws may have severe financial or other penalties for companies who have been found to have acted in a discrimnatory manner, for example, having software that has some form of {{bias}}.   This is parricularly probematic for compex AI algorithms and machine learning such as {{deep neural networks}} as it may be unclear what aspects are being used to make a decision.  Note that irrespectuve of the law in the nation(s) in which you are working or deploying systems, you should consider the {{ethical}} consequences of systems you produce, for example, bias for non-protected characteristcs such  social background, or wider {{societal impact of AI}}.
        )

    [antecedent-driven reasoning] =&gt; Array
        (
            [slug] =&gt; antecedent-driven reasoning
            [parent] =&gt; 
            [name] =&gt; antecedent-driven reasoning
            [canonical] =&gt; 
            [see] =&gt; bottom-up reasoning
            [description] =&gt; 
        )

    [apache hadoop] =&gt; Array
        (
            [slug] =&gt; apache hadoop
            [parent] =&gt; 
            [name] =&gt; Apache Hadoop
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; Hadoop is the Apache {{distributed computation}} system to allow efficient large scale processing of big data.  It includes an open source implementation of {{MapReduce}}.
        )

    [aperture] =&gt; Array
        (
            [slug] =&gt; aperture
            [parent] =&gt; 
            [name] =&gt; aperture
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; The aperture of a camera lens is the size of the hole through whuch light enters the camera.  This controls the amount of light entering the camera, so has to be larger in darker conditions.  It also affects the {{depth of field}} (what is in focus simultaneously) as the outer areas of the lens have more distortion, so that a smaller apperture leads to sharper images even if they are not at the exact {{focal point}} ... and also, becasue your pupil acts as apperture control in your eye, the reason why it is easier to read in bright light.
        )

    [appropriate intelligence] =&gt; Array
        (
            [slug] =&gt; appropriate intelligence
            [parent] =&gt; 
            [name] =&gt; appropriate intelligence
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; Appropriate intelligence is a deisgn heuristic for {{intelligent user interfaces}}.   When creating an ntelligent interface the temptation is to focus on making the AI as good as possible, but it is usually more importnt to make sure that the inetraction deisgned aorund the AI makes an overall system that works well.  In particular, it is important that when the AI fails (which it will!) this does not overly inconvenience the user.  Positive examples incude the Microsoft SUM (&amp;Sigma;) button and the design of {{onCue}}, whereas {{Clippy}} is an example of poor use of intelligence. as it interupted the user's workflow.
        )

    [appropriating technology] =&gt; Array
        (
            [slug] =&gt; appropriating technology
            [parent] =&gt; 
            [name] =&gt; appropriating technology
            [canonical] =&gt; 
            [see] =&gt; appropriation
            [description] =&gt; 
        )

    [appropriation] =&gt; Array
        (
            [slug] =&gt; appropriation
            [parent] =&gt; 
            [name] =&gt; appropriation
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; We often use objects and tools for purposes far different than their intended use, for example, cutting parcel tape with a key or proping open a door with a heavy book.  This process of appropriation can be used for technology, for example colouring cells in a spreadsheet to act as a rough image.
        )

    [architecture] =&gt; Array
        (
            [slug] =&gt; architecture
            [parent] =&gt; 
            [name] =&gt; architecture
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; Within computing architecture usually refers to the way in which software is structured.  Within AI this may include {{blackboard architectures}} and the choice of the types and sizes of the different layers in a {{deep neural network}}.  In addition a {{cognitive arcitecture}} is a way of representing the main processes and units in the human (or other animal) brain, but at a much higher level than individual {{neorones}}.
        )

    [area framework] =&gt; Array
        (
            [slug] =&gt; area framework
            [parent] =&gt; 
            [name] =&gt; AREA framework
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; The AREA framework hase been developed by EPSRC, the UK funding council that covers AI and computing, in order to promote {{responsible innovation}}.  The framework encourages researchers to consiser theor work under four headings: Anticipate, Reflect, Engage, and Act, each of which has further levels of advice.
        )

    [area under the curve] =&gt; Array
        (
            [slug] =&gt; area under the curve
            [parent] =&gt; 
            [name] =&gt; area under the curve
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; The area under the curve is an {{accuracy}} measure and is literally the area under the {{ROC}} curve.  If a decision rule has higher {{precision}} or higher {{recall}} it ends up with a larger area under the curve.  A {{random classifer}} has an area under the curve of 0.5, wheras a {{perfect classifier}} with 100% {{precsion}} and 100% {{recall}} has an area undet the curve of 1.  In general, because of the {{precision-recall trade-off}} the value lies somewhere between the two.
        )

    [arithmetic mean] =&gt; Array
        (
            [slug] =&gt; arithmetic mean
            [parent] =&gt; 
            [name] =&gt; arithmetic mean
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; The arithmetic mean (often simply the mean) adds up the values of a numercal feature ina dataset, for example, the hights of people ina. room, and divides by the total number of indviduals.  For example, the mean of the numbers [1,7,6,2,4] is (1+7+6+2+4)/5 = 20/5 = 4.  The term {{average}} may refer to the arithmetic mean or may refer to the {{median}}.  The {{median}} is often a better {{centrality}} measure than the mean as it is more stable espedcially given extreme values, or for {{non-Normal distributions}} such as the {{power law distributions}}.  However, the mean has nice mathematical properties that make it easier to process, for example, if you have caclualted the mean of two datasets (say m&lt;sub&gt;1&lt;/sub&gt; and m&lt;sub&gt;2&lt;/sub&gt;) and know how many in each  (say N&lt;sub&gt;1&lt;/sub&gt; and N&lt;sub&gt;2&lt;/sub&gt;) it is possible to combine these: &lt;br&gt; overall mean = (m&lt;sub&gt;1&lt;/sub&gt; x N&lt;sub&gt;1&lt;/sub&gt; + m&lt;sub&gt;2&lt;/sub&gt; x N&lt;sub&gt;2&lt;/sub&gt;) / (N&lt;sub&gt;1&lt;/sub&gt; + N&lt;sub&gt;2&lt;/sub&gt;)
        )

    [arma] =&gt; Array
        (
            [slug] =&gt; arma
            [parent] =&gt; 
            [name] =&gt; ARMA (Auto Regressive Moving Average)
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; ARMA models use a combination of {{auto-regressive}} and {{moving average}} techniques to model time series.  It is a form of {{linear model}} for stochastic processes.
        )

    [arthur c. clarke] =&gt; Array
        (
            [slug] =&gt; arthur c. clarke
            [parent] =&gt; 
            [name] =&gt; Arthur C. Clarke
            [canonical] =&gt; clarke, arthur c
            [see] =&gt; 
            [description] =&gt; 
        )

    [artificial general intelligence] =&gt; Array
        (
            [slug] =&gt; artificial general intelligence
            [parent] =&gt; 
            [name] =&gt; artificial general intelligence
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; Artificial general intelligence (AGI) is the idea or goal of proecuing AI that can be applied across all, or at least a wide range, of domains and problems.  Early AI systems tended to be very specific in the kinds of problems they addressed based partly on their mode of {{reasoning}} and partly on their {{knowledge base}}.  The use of {{deep leaerning}} and in particular {{large-language models}} has rasied the possibility that these systems, currently relying on large volumes of himan generated materials, may already be at the edge of AGI.   Others disagree, especially as these systems are effectively reaching the limits of human-generated data and will have to rely on some form of {[self learning}}.
        )

    [artificial intelligence] =&gt; Array
        (
            [slug] =&gt; artificial intelligence
            [parent] =&gt; 
            [name] =&gt; artificial intelligence
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; Artificial intelligence has many definitions, but is broadly about computers that do things which, if performed by a human, would be considered intelligent.   Proponents of {{strong AI}} believe that it  will ultimately be possible to create machine intelligence that is equivalent in all sense to a human; whereas others take a {{weak AI}} approach that is happy to see inspiration from human intelligence realised in machines.  A more pragmatic approach simply looks for algorithms and techniques that can be used in practical applications.
        )

    [artificial intelligence!definition] =&gt; Array
        (
            [slug] =&gt; artificial intelligence!definition
            [parent] =&gt; artificial intelligence
            [name] =&gt; definition
            [canonical] =&gt; artificial intelligence
            [see] =&gt; 
            [description] =&gt; 
        )

    [artificial life] =&gt; Array
        (
            [slug] =&gt; artificial life
            [parent] =&gt; 
            [name] =&gt; artificial life
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; The term artificial life refers to a range of types of model, from those that operate are a level closer to baterium to human socities.  A common low-level exmaple are {{celluar automota}} such as  {{Conway}}'s {{Game of Life}}.  Higher level examples include models of {{flocking behaviour}} or modelling of human movements in emergency planning.  {{Swarm computing}} can also be seen as a form of artificial life. &lt;br&gt;&lt;br&gt; Artificial life can alos refer to the design of biological matter, often using AI to drive the process.  This includes {{xenobots}}, groups of cells desitned to act like living robots; and even completely designing DNA from scratch.
        )

    [artificial neural networks] =&gt; Array
        (
            [slug] =&gt; artificial neural networks
            [parent] =&gt; 
            [name] =&gt; artificial neural networks
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; Artificial neural networks is another term for computer-based {{neural networks}}, largely used when one wants to distinguish them from biological neural networks in the brain.  This is most likely to be the case in more fundamental work where artficial networks are being used in neuroscience to understand brain behaviour or where novel algorithms are being developed based on brain neural analogies.  In areas where the digital nature is clear the term {{neural network}} is used.
        )

    [artificial society] =&gt; Array
        (
            [slug] =&gt; artificial society
            [parent] =&gt; 
            [name] =&gt; artificial society
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; Artificial societies are {{agent-based models}} to emulate social phenomena, current and hsitorical, for example neolithic migration patterns.  Relatively simple models of individuals often give rise to overall behaviours that emulate actual social processes, rather like {{flocking modles}} do for animal movements.
        )

    [asimov, isaac] =&gt; Array
        (
            [slug] =&gt; asimov, isaac
            [parent] =&gt; 
            [name] =&gt; Asimov, Isaac
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; Science fiction author famous for his {{three laws of robotics}}.
        )

    [aspect ratio] =&gt; Array
        (
            [slug] =&gt; aspect ratio
            [parent] =&gt; 
            [name] =&gt; aspect ratio
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; For images such as photographs, this is the ratio between the width and height of the image.  For example a square image has an aspect artion of 1:1 whereas a traditional photograph is 4:3, that is a 1/3 wider than it is high.   For a digital image the aspect ratio usually refers to the ratio of the number of pixels, for example a 1024x768 has a 4:3 aspect ration, wheas a wide screen at 1920x1080 has an aspect ration of 16:9.  Most commoly each pixel is square, and this is certainly the case on almost all computer screens, but occasionally in specialised applications pixels may be rectangular an have an asoect ration that is not 1:1; when this is the case the apect artio of the overall inage and the pixel rations wouydnot be the same, for example a 1000x500 image where each pixel had a 3:2 aspect ration would have an overall aspect ratio of 3:1 when printed/viewed.
        )

    [associative memory] =&gt; Array
        (
            [slug] =&gt; associative memory
            [parent] =&gt; 
            [name] =&gt; associative memory
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; An assocative memory is one which tries to learn relationshop between an imput and an output so that when given an unseen input it can allocate an appropriate output.  In some case the output is very different form the output. For example, where an image is associated with a class.  In others the memory can be used as an {{auto-encoder}} where the input and output are identical -- this form of {{auto-associative memory}} initially does not seem useful, however it may be used where the input is in some way incomplete or noisy and the memory recreates a full or noise-free image.
        )

    [asymmetric distribution] =&gt; Array
        (
            [slug] =&gt; asymmetric distribution
            [parent] =&gt; 
            [name] =&gt; asymmetric distribution
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; An asymmetric distribution is a {{probability distrubution}} that is asymetric about its {{median}}. For example, waiting times in many contexts are often highly asymetric, not least because ther cannot be negative waiting times, but aso becasue there may also be occasional long waiting times.   Asymmetric distributions are often heavily {{skewed}}, as in the example of waiting times, with a {{long tail}} on one side, but may also simply be different shapes.
        )

    [attention mechanisms] =&gt; Array
        (
            [slug] =&gt; attention mechanisms
            [parent] =&gt; 
            [name] =&gt; attention mechanisms
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; Attention mechanisms are used in {{machine learning}}, particularly for text and in {{large language models}} and other forms of {{deep learning}}.  The concept borrows form human attention, which is critical for cognition.  Given a sequence of tokens (e.g. words in a text), simple {{window-based methods}} would treat all past tokens in the window as equally relevant for predicting or otherwise leaening based on a current token.  In contrast attention mechanisms attempt to work out past tokens that are especially relevant and gves these higher weight (attention) during training.  This may be acheived by creating an interest vector for each input (as part of the {{deep learning}} process) that in some way represents its topic area and then matching the interest vector of past tokens with the current token.  Some attention mechanisms will also retain past tokens that have proved especially relevant based on their relevance to later tokens, thus allowing memory beyond the window.  {{Salience}} can be combined with attention mechanisms by rating the unusuallness of tokens and token sub-seqences and use these to weight tokens together with similarity to the current token.
        )

    [augmented transition network] =&gt; Array
        (
            [slug] =&gt; augmented transition network
            [parent] =&gt; 
            [name] =&gt; augmented transition network
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; An augmented transition network is a {{transition network}} that allows procedures to be attached to arcs that act as an additional test to allow the arc to be traversed.  In particular this offers a mechanism to model {{context sensitive grammars}} and hence more complex patterns of input than would be possible with a simple {{transition network}}.
        )

    [augmented transition networks] =&gt; Array
        (
            [slug] =&gt; augmented transition networks
            [parent] =&gt; 
            [name] =&gt; augmented transition networks
            [canonical] =&gt; augmented transition network
            [see] =&gt; 
            [description] =&gt; 
        )

    [authority file] =&gt; Array
        (
            [slug] =&gt; authority file
            [parent] =&gt; 
            [name] =&gt; authority file
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; An authority file is used to ensure that named people, places or other things can be connected in a dataset.  For example, the name 'Alan Dix' might refer to the author of AI, HCI and statistics textbooks, or Alan Dix the theatre maker.  An authrity file would allocate each a unique identifier, say AlanDix_276493 and AlanDix_835672 and then these identifiers would be used in orther data records to ensure it is clear whcih Alan Dix is being fereed to.  In {{pseudonomised data}} the identifeir is deliberately chosen to be anonymous (unlike those in the example) and the authorty file is either kept securely or deleted after use.
        )

    [auto-associative] =&gt; Array
        (
            [slug] =&gt; auto-associative
            [parent] =&gt; 
            [name] =&gt; auto-associative
            [canonical] =&gt; auto-associative memory
            [see] =&gt; 
            [description] =&gt; 
        )

    [auto-associative memory] =&gt; Array
        (
            [slug] =&gt; auto-associative memory
            [parent] =&gt; 
            [name] =&gt; auto-associative memory
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; An auto-associative memory is an {{assocative memory}} where the input and output are identical.  This initially does not seem useful, however it may be used where the input is in some way incomplete or noisy and the {{auto-encoder}} recreates a full or noise-free image.
        )

    [auto-regressive] =&gt; Array
        (
            [slug] =&gt; auto-regressive
            [parent] =&gt; 
            [name] =&gt; auto-regressive
            [canonical] =&gt; auto-regressive model
            [see] =&gt; 
            [description] =&gt; 
        )

    [auto-regressive model] =&gt; Array
        (
            [slug] =&gt; auto-regressive model
            [parent] =&gt; 
            [name] =&gt; auto-regressive model
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; An auto-regressive model (of order n) is used when a time series is believed to be of the form x&lt;sub&gt;t&lt;/sub&gt; =  a&lt;sub&gt;1&lt;/sub&gt; x&lt;sub&gt;t-1&lt;/sub&gt; + &hellip; + a&lt;sub&gt;1&lt;/sub&gt; x&lt;sub&gt;t-(n-1)&lt;/sub&gt; + noise&lt;sub&gt;t&lt;/sub&gt;.  The simplest example is a {{drunkards walk}} where n=1 and a&lt;sub&gt;1&lt;/sub&gt;=1.
        )

    [auto-regressive models] =&gt; Array
        (
            [slug] =&gt; auto-regressive models
            [parent] =&gt; 
            [name] =&gt; auto-regressive models
            [canonical] =&gt; auto-regressive model
            [see] =&gt; 
            [description] =&gt; 
        )

    [auto-regressive moving average] =&gt; Array
        (
            [slug] =&gt; auto-regressive moving average
            [parent] =&gt; 
            [name] =&gt; auto-regressive moving average
            [canonical] =&gt; arma
            [see] =&gt; 
            [description] =&gt; 
        )

    [autoassociative memory] =&gt; Array
        (
            [slug] =&gt; autoassociative memory
            [parent] =&gt; 
            [name] =&gt; autoassociative memory
            [canonical] =&gt; auto-associative memory
            [see] =&gt; 
            [description] =&gt; 
        )

    [autoencoder] =&gt; Array
        (
            [slug] =&gt; autoencoder
            [parent] =&gt; 
            [name] =&gt; autoencoder
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; Autoencoder is another word for {{autoassocitve memory}}, but more often used in recent {{neural network}} literature.  In the training data the input and output are identical, often the same image as each other.   The network thius learns to associate the image with itself.  This initially does not seem very intersting except that internal layers typically are substantially smaller than the full input.  A small middle layer is effectvely coding the key features of the image for reconstruction.  Becasue of this the inner, {{pinch point}} layer can sometimes be used as a form of {{data-reduction}} for other knds of network, for example a classifier.  Alternatively the network may be able to recreate images from partial inputs or even generate new ones by setting the inner layer.
        )

    [autoencoders] =&gt; Array
        (
            [slug] =&gt; autoencoders
            [parent] =&gt; 
            [name] =&gt; autoencoders
            [canonical] =&gt; autoencoder
            [see] =&gt; 
            [description] =&gt; 
        )

    [automated decision] =&gt; Array
        (
            [slug] =&gt; automated decision
            [parent] =&gt; 
            [name] =&gt; automated decision
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; An automated decsion is one where the final decsion is made by the computer system without a human having the final say.  For sensitive decsions such as medical or military applications, many prefer there to be a {{human in the loop}} at very least to veto or put a check on the automated decisions.  Humans are almost always still involved in fully automated, either in supplying data or executing the actions.  For example, in applications such as Uber or Lyft the descison about which car is sent to which client is completely automated even though humans drive the cars.
        )

    [automation bias] =&gt; Array
        (
            [slug] =&gt; automation bias
            [parent] =&gt; 
            [name] =&gt; automation bias
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; Automation bias is the tendency for humans to agree with computer advice or be swayed by computational data even when the automated system.  This can be problematic when the system fails, is error prone or doesn't take into account all aspects of a situation.   For example, an AI system may use a small number of features to advise on whether someone is at high or low risk when they enter an hospital emergnecy department; if the person is obviously not breathing the paramedic or nurse shoudl clearly igbore this, but there can be a tendency to assume the computer knows best.  This can also lead to {{bias}} for example, the {COMPAS}} system that has been used by judges as part of parole decisions in the US has been criticsed as embodying elements of racial bias.  ULtimately the judge makes the decsion, but may be swayed by the system's assessment.
        )

    [autonomous car] =&gt; Array
        (
            [slug] =&gt; autonomous car
            [parent] =&gt; 
            [name] =&gt; autonomous car
            [canonical] =&gt; autonomous car
            [see] =&gt; 
            [description] =&gt; 
        )

    [autonomous vehicle] =&gt; Array
        (
            [slug] =&gt; autonomous vehicle
            [parent] =&gt; 
            [name] =&gt; autonomous vehicle
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; An autonomous vehicle is one that partially or fully drives itself.   In a semi-autonmous mode, this may aid particular manouvres such as parallel parking, or driving within lane markings on a long drive.  In fully autonomous mode, the car also has to deal with ever changing road conditons; other road users including pedestrians and other vehicles; and hazards suhc aas child or animal running onto the road. This usually involves a combination of vision and other sensors linked to rapid response planning and hazard avoidance.   
        )

    [autonomous vehicles] =&gt; Array
        (
            [slug] =&gt; autonomous vehicles
            [parent] =&gt; 
            [name] =&gt; autonomous vehicles
            [canonical] =&gt; autonomous vehicles
            [see] =&gt; 
            [description] =&gt; 
        )

    [autonomous weapons] =&gt; Array
        (
            [slug] =&gt; autonomous weapons
            [parent] =&gt; 
            [name] =&gt; autonomous weapons
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; An autonomous weapon is any form of missile, drone or armed vehicle that uses AI to make its own decsions.  This may be about navigating terrain, but may also include choosing targets.  The latter has given rise to substantial concern as there is no {{human in the loop}} to  ensure that only acceptable military targets are attacked.  AI is used extensively in other military applications, including offering advice on targetting, but without the systems being completely autonomous.
        )

    [astar algorithm] =&gt; Array
        (
            [slug] =&gt; astar algorithm
            [parent] =&gt; 
            [name] =&gt; A&lt;sup&gt;*&lt;/sup&gt; algorithm
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; The A&lt;sup&gt;*&lt;/sup&gt; algorithm is a {{heuristic search}} most often used for finding shortest-path routes.  If the {{heuristic}} guarantees a {{lower bound}} on the costs of any node below, then this can be used to {{prune}} whole branches where the heurisic exceeds the current best solution.  That is, it can be seen as heuristic version of {{branch and bound}}.  Often the raw {{heauristic}} is about the path beyond the node, for example the crows-fly distance to the target location in geographic {{shortest path}} search, in which case the distance along the path to the node has to be added to the raw heuristic to create a total path {{lower bound}}.  The heauristic can also be used to guide the next node to explore in a variant of {{hill climbing}}.
        )

    [a${}^\ast$ algorithm] =&gt; Array
        (
            [slug] =&gt; a${}^\ast$ algorithm
            [parent] =&gt; 
            [name] =&gt; a${}^\ast$ algorithm
            [canonical] =&gt; astar algorithm
            [see] =&gt; 
            [description] =&gt; 
        )

    [babbage] =&gt; Array
        (
            [slug] =&gt; babbage
            [parent] =&gt; 
            [name] =&gt; babbage
            [canonical] =&gt; babbage-charles
            [see] =&gt; 
            [description] =&gt; 
        )

    [babbage-charles] =&gt; Array
        (
            [slug] =&gt; babbage-charles
            [parent] =&gt; 
            [name] =&gt; Babbage, Charles
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; Charles Babbage was an nineteenth century inventor, famous for his {{Difference Engine}}, which used mechanical means to compute complex calculations, and is thus seen as one of the pioneers of computing.  He later invdented a more complex programmable device, the {{Analytic Engine}}, which is regarded as the first programmable computer, with his collaborator, {{Ada Lovelace}}, as the first programmer.
        )

    [backprop] =&gt; Array
        (
            [slug] =&gt; backprop
            [parent] =&gt; 
            [name] =&gt; backprop
            [canonical] =&gt; backpropagation
            [see] =&gt; 
            [description] =&gt; 
        )

    [backpropagation] =&gt; Array
        (
            [slug] =&gt; backpropagation
            [parent] =&gt; 
            [name] =&gt; backpropagation
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; Backpropogation is a {{supervised learning}}  for multi-level {{neural networks}}.   For each training example, it takes the difference between the expected and actual output at the final layer and then uses the {{differential}} of the {sigmoid}}  function at each node to work out error values at earlier layers and hence update the weights on links between nodes.
        )

    [backtracking] =&gt; Array
        (
            [slug] =&gt; backtracking
            [parent] =&gt; 
            [name] =&gt; backtracking
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; Backtracking is when a {{search}} or related algorithm returns to previous state.  This maybe because the algorithm has got stuck, for example when a {{tree search}} encounters a node that is {{infeasible}}, that is does not satisfy all of the {{constraints}} for an acceptable solution.  Alternatively, when the algorithm has found some form of a soluution, it may want to look for others.
        )

    [backward branching factor] =&gt; Array
        (
            [slug] =&gt; backward branching factor
            [parent] =&gt; 
            [name] =&gt; backward branching factor
            [canonical] =&gt; 
            [see] =&gt; branching factor
            [description] =&gt; 
        )

    [backward chaining] =&gt; Array
        (
            [slug] =&gt; backward chaining
            [parent] =&gt; 
            [name] =&gt; backward chaining
            [canonical] =&gt; 
            [see] =&gt; backward reasoning
            [description] =&gt; 
        )

    [backward reasoning] =&gt; Array
        (
            [slug] =&gt; backward reasoning
            [parent] =&gt; 
            [name] =&gt; backward reasoning
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; Backward reasoning is used in {{reasoning systems}} when rules are run'backwards' in order to ask &quot;what would need to be true in order to show X&quot;. That is, given a goal to find sub-goals that need to be solved, themsleves using backward reasoning.  It is particularly useful when the backward {{branching factor}} is smaller than the forwards {{branching factor}}, for example wgere there are huge numbers of fact known combing them to generate derived knowledge through {{forward reasoning}} would generate vast numbers of anticident facts in the hope that one of them iis useful.  In contrast only a small number of them will be relevant to particular goal.  In an {{expert system}} backward reasoning can be used to suggest questiosn to ask; for example, given &quot;all men are mortal&quot;, and the goal is to prove &quot;Socrates is mortal&quot;, a backward reasoning system might ask &quot;Is Socrates a man?&quot;
        )

    [bag of words] =&gt; Array
        (
            [slug] =&gt; bag of words
            [parent] =&gt; 
            [name] =&gt; bag of words
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; A document consists of structured text: sentances, paragaphs, and meaningful phrases.  Bag of words techniques ignore this structure reducing the document to the set of words with a frequency count for each.  This is used for {{similarity metrics}} such as the {{Jaccard similarity}} and {{cosine similarity}}.
        )

    [bard] =&gt; Array
        (
            [slug] =&gt; bard
            [parent] =&gt; 
            [name] =&gt; Bard
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; Bard is a {{Google}} {{chatbot}} based on a {{large language model}}, a competitor to ChatGPT.
        )

    [base of the logarithm] =&gt; Array
        (
            [slug] =&gt; base of the logarithm
            [parent] =&gt; 
            [name] =&gt; base of the logarithm
            [canonical] =&gt; logarithm base
            [see] =&gt; 
            [description] =&gt; 
        )

    [base rate] =&gt; Array
        (
            [slug] =&gt; base rate
            [parent] =&gt; 
            [name] =&gt; base rate
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; The base rate is the underlying frequency or proability of a phenomena.  For example, if one in thousand people are likely to be suffering from flu during the windter, then the base rate of flu is 0.001.  The base rate is critical when using evidence to make diagnoses or ither decisions.  For example, if flu and the common cold have similar symptoms and ther base rate of the common cold is one in 50, then someone with a runny nose and headache is more likely to be suffering form a cold than flu.  This is captured formally in {{Bayes theorem}}.  &lt;br&gt;&lt;.br&gt;Base rates are also important when trying to reduce {{bias}} or {{discrimination}} as we try to be fair to an individual irrespectve of the base rate for their demographic.  Without taking remedial action data-based {{machine learning}} will tend to be discriminatory becasue of this, even if the data is free of human bias.
        )

    [base rates] =&gt; Array
        (
            [slug] =&gt; base rates
            [parent] =&gt; 
            [name] =&gt; base rates
            [canonical] =&gt; base rate
            [see] =&gt; 
            [description] =&gt; 
        )

    [bayes theorem] =&gt; Array
        (
            [slug] =&gt; bayes theorem
            [parent] =&gt; 
            [name] =&gt; Bayes Theorem
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; Bayes theorem is a part of {{probability}} theory that relates conditional probabilities and can be used to update beliefs (expressed as probabilities) to be updated based on new evidence.  This is used in a number of approaches in AI including {{Bayesian networks}}.  To be precise Bayes theorem says: &lt;br&gt;. P( X | E ) = ( P( E | X ) P(X) ) / sum ( P( E | X&lt;sub&gt;i&lt;/sub&gt; ) P(X&lt;sub&gt;i&lt;/sub&gt;) ) &lt;br&gt; where X is some unknown state (such as &quot;it is raining&quot;) the sum is takend over all possible states (e.g. &quot;it is dry&quot;, &quot;it is snowing&quot;) and E is some other observable (often interpreted as evidence, such as &quot;there's a wet umbrella in the hall-stand&quot;).  It effectively allows one to turn condirtonal probabiliies round.  When being used to update beliefs using evidence the probabilities  P(X&lt;sub&gt;i&lt;/sub&gt;) are known as the {{prior distribution}}, whereas P(X&lt;sub&gt;j&lt;/sub&gt;|E) is the {{posterior distribution}.
        )

    [bayesian methods] =&gt; Array
        (
            [slug] =&gt; bayesian methods
            [parent] =&gt; 
            [name] =&gt; Bayesian methods
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; This refers to any kinds of methods that use {{Bayes theorem}}.
        )

    [bayesian network] =&gt; Array
        (
            [slug] =&gt; bayesian network
            [parent] =&gt; 
            [name] =&gt; Bayesian network
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; A Bayesian network is a {{network}} of evidence and deriveed properties/facts linked by {{conditional probabilities}} that can be used to work out the probabilities of various facts given evidence. The same network can typically be used to make inferences in different dircetions, for example working out the probability of seeing someone with an umbrella if it is rainy, or the probability that it is rainy if you spot someone with an umbrella.
        )

    [bayesian networks] =&gt; Array
        (
            [slug] =&gt; bayesian networks
            [parent] =&gt; 
            [name] =&gt; Bayesian networks
            [canonical] =&gt; bayesian network
            [see] =&gt; 
            [description] =&gt; 
        )

    [bayesian reasoning] =&gt; Array
        (
            [slug] =&gt; bayesian reasoning
            [parent] =&gt; 
            [name] =&gt; Bayesian reasoning
            [canonical] =&gt; bayes theorem
            [see] =&gt; 
            [description] =&gt; 
        )

    [berners-lee] =&gt; Array
        (
            [slug] =&gt; berners-lee
            [parent] =&gt; 
            [name] =&gt; berners-lee
            [canonical] =&gt; berners-lee, tim
            [see] =&gt; 
            [description] =&gt; 
        )

    [berners-lee, tim] =&gt; Array
        (
            [slug] =&gt; berners-lee, tim
            [parent] =&gt; 
            [name] =&gt; Berners-Lee, Tim
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; Tim Berners-Lee is the inventor of the {{World-Wide Web}}, which he developed while working at {{CERN}} initially as a means of sharing scientific data.
        )

    [best first] =&gt; Array
        (
            [slug] =&gt; best first
            [parent] =&gt; 
            [name] =&gt; best first
            [canonical] =&gt; best first search
            [see] =&gt; 
            [description] =&gt; 
        )

    [best first search] =&gt; Array
        (
            [slug] =&gt; best first search
            [parent] =&gt; 
            [name] =&gt; best first search
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; Best-first search is a form of {{tree search}} where there is a {{heuristic evaluation}} for each node. At each step, the best-first algorithm chooses the node with the best heuristic to follow down the tree, only {{backtracking}} when it gets to a leaf node, or in some other way reaches em impasse.
        )

    [bias] =&gt; Array
        (
            [slug] =&gt; bias
            [parent] =&gt; 
            [name] =&gt; bias
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; A system has bias if it unfairly benfits or is more positive to one gorup of people compared with another.   Bias can arise for many reasons (see {{chap:bias}}) including when the training data in some way embodies human discrimination; this may happen even if the algorithm appears to be being fair.  Bias is diffidult to avoid entirely as it depends on what we consider fair characetrstics, for example, job recruitment might quite reasonably use exam results of relevant subjects, so is in some ways biased against those who, for no fault of their own , have not had as good an education.  What is deemed acceptable and avodiing unacceptable bias is in part a personal and corporate ethical decsion, but is also a legal one as many nations have {{anti-discrimination laws}}.    Note the term bias also has a very precise statsitsical meaning relating to the whether or not the long-term output of a statistical measure of some phenomenon is equal to the true value.
        )

    [biasing] =&gt; Array
        (
            [slug] =&gt; biasing
            [parent] =&gt; 
            [name] =&gt; biasing
            [canonical] =&gt; bias
            [see] =&gt; 
            [description] =&gt; 
        )

    [bibliographic database] =&gt; Array
        (
            [slug] =&gt; bibliographic database
            [parent] =&gt; 
            [name] =&gt; bibliographic database
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; 
        )

    [bibliographic databases] =&gt; Array
        (
            [slug] =&gt; bibliographic databases
            [parent] =&gt; 
            [name] =&gt; bibliographic databases
            [canonical] =&gt; bibliographic database
            [see] =&gt; 
            [description] =&gt; 
        )

    [big blue] =&gt; Array
        (
            [slug] =&gt; big blue
            [parent] =&gt; 
            [name] =&gt; Big Blue
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; IBM Big Blue is the chess program that defeated the grandmaster Gary Kasprov in 1997.  This was a shock at the time as chess at grandmaster was considered too difficult for a computer.  Deep Blue made extensive use of 'brute force'  {{lookahead}} running on an IBM supercomputer and Kasprov labelled it an 'alien opponent' as it played in a very different way to a human.
        )

    [big data] =&gt; Array
        (
            [slug] =&gt; big data
            [parent] =&gt; 
            [name] =&gt; big data
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; 
        )

    [binary image] =&gt; Array
        (
            [slug] =&gt; binary image
            [parent] =&gt; 
            [name] =&gt; binary image
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; A binary image is a black-and-white {{image}} where the {{pixels}} only take the values 1 or 0.
        )

    [bitmap image] =&gt; Array
        (
            [slug] =&gt; bitmap image
            [parent] =&gt; 
            [name] =&gt; bitmap image
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; A bitmap image represents a visual scene or {{image}} as a matrix of {{pixesl}} each of which is a binary representation of a colour.  This may be direct, for example encoding red, blue and green levels each as 8 bit values, or may be indexed, wheer each pixel is an index into a colour map.  The latter is less common in modern formats.  Often bitmap images are {{compressed}} so that the storage format does not driectly represent the {{pixels}}, but these can be restored for display or processing.
        )

    [blackboard architecture] =&gt; Array
        (
            [slug] =&gt; blackboard architecture
            [parent] =&gt; 
            [name] =&gt; blackboard architecture
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; Blackboard architectures are used in {{agent}}-based AI systems to share information between {{agents}}.  the blackboard is a gared information structure avaiable to all the agents.  Each agent mnitors the blackboard and when it spots soemthing ot knows how towrk with takes the elemnt, processs it and then posts is results back onto the blackbiard in the hope that anoter agent will pick it up.   &lt;br&gt;&lt;br&gt;As an example, consider a maths solving AI system.  There is one agent (call it A) that knows about basic adding up, another (call it B) that knows how group things into brackets, one (call it S) that can swop sides, M that ca multiple/divide sides by the sam thing, and others still for other kinds of manipulation. Initially the blackbaord has the equation &quot;3x = x+6&quot;.  When agent A looks at it, it can only do number arithmetic, similarly B can do nothing, but agent S can process it and posst back &quot;3x &ndash; x = 6&quot;.  At this point B kicks in and posts back &quot;(3&ndash;1)x = 6&quot;.  A now has some basic aritjmatic so posts back &quot;2x = 6&quot;, M sees it and posts &quot;x = 6/2&quot;, finally agent A posts &quot;x = 3&quot;, the solution.&lt;br&gt;&lt;br&gt;In practice agent archotectures will have various machnaicms to make this process more efficient such as ways for an agent to register patterns that are of interest to avoid constant {{polling}} of the blackboard; there may also be some sort of coordinating agent, that helps to prioritise the agents.  
        )

    [blake lemoine] =&gt; Array
        (
            [slug] =&gt; blake lemoine
            [parent] =&gt; 
            [name] =&gt; blake lemoine
            [canonical] =&gt; lemoine, blake
            [see] =&gt; 
            [description] =&gt; 
        )

    [boden-margaret] =&gt; Array
        (
            [slug] =&gt; boden-margaret
            [parent] =&gt; 
            [name] =&gt; Boden, Margaret
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; 
        )

    [boden, m] =&gt; Array
        (
            [slug] =&gt; boden, m
            [parent] =&gt; 
            [name] =&gt; boden, m
            [canonical] =&gt; boden-margaret
            [see] =&gt; 
            [description] =&gt; 
        )

    [boltzmann machine] =&gt; Array
        (
            [slug] =&gt; boltzmann machine
            [parent] =&gt; 
            [name] =&gt; Boltzmann machine
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; The Boltzmann machine is an early form of {{neural network}} based on {{spin glass models}} in physics, which also have parallels in human brain mechanisms.   Boltzman machines are often used as an {{autoencoder}}.  Boltzman machines have two types of nodes (or layers): visible (which can be used for input and output) and hidden (which often act as low-demiensional encodings of inputs).  In a standard Boltzmann machine all nodes are fully interconnected, but in {{restricted Boltzmann machines}} there are only connections between the layers.
        )

    [boltzmann machines] =&gt; Array
        (
            [slug] =&gt; boltzmann machines
            [parent] =&gt; 
            [name] =&gt; boltzmann machines
            [canonical] =&gt; boltzmann machine
            [see] =&gt; 
            [description] =&gt; 
        )

    [bombe] =&gt; Array
        (
            [slug] =&gt; bombe
            [parent] =&gt; 
            [name] =&gt; Bombe
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; The Bombe was a computational device designed by {{Alan Turing}} and others at Bletchley to help decrypt {{Enigma machine}} messages.  This was one of the inspiratons for his later work studying computation.
        )

    [boolean network] =&gt; Array
        (
            [slug] =&gt; boolean network
            [parent] =&gt; 
            [name] =&gt; Boolean network
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; A Boolean network is a form of neural netwrok where all of the values are Boolean (true/false) and the internal nodes logic functions such as AND or NOR.  It is can be hard to train a Boolean network as they are discontinuous, but they can be very efficient to run, especially if compiled into a {{FPGA}}.  In some case more compelx {{neural netowrks}} can be simplified into Boolean networks.
        )

    [boosting] =&gt; Array
        (
            [slug] =&gt; boosting
            [parent] =&gt; 
            [name] =&gt; boosting
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; 
        )

    [bootstrapped] =&gt; Array
        (
            [slug] =&gt; bootstrapped
            [parent] =&gt; 
            [name] =&gt; bootstrapped
            [canonical] =&gt; bootstrapping
            [see] =&gt; 
            [description] =&gt; 
        )

    [bootstrapping] =&gt; Array
        (
            [slug] =&gt; bootstrapping
            [parent] =&gt; 
            [name] =&gt; bootstrapping
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; 
        )

    [bottom-up algorithm] =&gt; Array
        (
            [slug] =&gt; bottom-up algorithm
            [parent] =&gt; 
            [name] =&gt; bottom-up algorithm
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; 
        )

    [bottom-up algorithms] =&gt; Array
        (
            [slug] =&gt; bottom-up algorithms
            [parent] =&gt; 
            [name] =&gt; bottom-up algorithms
            [canonical] =&gt; bottom-up algorithm
            [see] =&gt; 
            [description] =&gt; 
        )

    [bottom-up reasoning] =&gt; Array
        (
            [slug] =&gt; bottom-up reasoning
            [parent] =&gt; 
            [name] =&gt; bottom-up reasoning
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; 
        )

    [boundary example] =&gt; Array
        (
            [slug] =&gt; boundary example
            [parent] =&gt; 
            [name] =&gt; boundary example
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; Boundary examples are taken from training data or generated and are at or near the boundary between classes.  It can be useful to identify boundary examples during {{machine learning}} as they are places where mis-classifications are likely and hence may be more heavily weighted to ensure accurate boundary behaviour.   Boundary examples can also be used as part of {{explainable AI}} in order to help a human understand classes created by an AI system.
        )

    [boundary objects] =&gt; Array
        (
            [slug] =&gt; boundary objects
            [parent] =&gt; 
            [name] =&gt; boundary objects
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; Boundary objects are physical, digital or conceptual items that can act as a point ofcommunication or collaboration between types of people who otherwsie may have very different goals, knowkedge and backgorunds.  The term was first coined to consider the various groups who work around zoological exhibits in a museum, and examples in that context include the phsyical exhibits and the texonomic structyre used to classify them.   As an example in a digital context, archirects and engineers might collaborate around a CAD drawing even though they understand diffent aspects of it, and use it for different purposes.  In {{synergistic}} AI-human collaboration, we might seek to create bondary objects that have meaning for both human and system.  For example, in {{Query-by-Browsing}} the selected items in the list and SQL query are used differently by AI and human, but are a point of connection between them.
        )

    [branch and bound] =&gt; Array
        (
            [slug] =&gt; branch and bound
            [parent] =&gt; 
            [name] =&gt; branch and bound
            [canonical] =&gt; branch and bound search
            [see] =&gt; 
            [description] =&gt; 
        )

    [branch and bound search] =&gt; Array
        (
            [slug] =&gt; branch and bound search
            [parent] =&gt; 
            [name] =&gt; branch and bound search
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; Branch and bound search is used in {{optimisation}} situations for {{tree seadch}} or {{{graph search}} when there is a cost associated with the path taken to reach a solution, and we want to find the solution with the least path cost.  The path cost is often {{additive}}, for example minimising the number of moves to get to a solution, distances between locations along a geographic route, where each move in a puzzle has a cost.  However, branch and bound search does not rely on this, merely that the costs of a path are {{monotonic}}, that isnthey increase if the path gets longer.  The key insight of branch and bound search is that once we have found best-yet solution, we can {{prune}} whole branches for which the path to the branch node exceeds the current best solution.  In addition, the path cost to a node can be used as a {{heuristic}} to order the {{open list}} and choose which nodes to explore first.
        )

    [branching factor] =&gt; Array
        (
            [slug] =&gt; branching factor
            [parent] =&gt; 
            [name] =&gt; branching factor
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; In a {{tree} the branching factor of a node refers to the number of branches/children of the node.  For the tree as a whole this may be fixed (e.g. a branching factor of 2 for a {{binary tree}}) or may vary between nodes (as is often the case in {{search trees}}).  One may therefore talk about a maximum or average branching factor of a tree.  For example, for the first move in {{Go}} there are 19x19 possible moves, so the branching factor at the root is 361 (although some are equivalent). &lt;br&gt;&lt;br&gt;. In a {{directed graph}} , that is when the links/arcs between nodes have a direction, one can talk about the forward and backward branching factors.  The former is the number of outward links and the latter the numer of inward links.
        )

    [branching factor!backward] =&gt; Array
        (
            [slug] =&gt; branching factor!backward
            [parent] =&gt; branching factor
            [name] =&gt; backward
            [canonical] =&gt; branching factor
            [see] =&gt; 
            [description] =&gt; 
        )

    [branching factor!forward] =&gt; Array
        (
            [slug] =&gt; branching factor!forward
            [parent] =&gt; branching factor
            [name] =&gt; forward
            [canonical] =&gt; branching factor
            [see] =&gt; 
            [description] =&gt; 
        )

    [breadth first] =&gt; Array
        (
            [slug] =&gt; breadth first
            [parent] =&gt; 
            [name] =&gt; breadth first
            [canonical] =&gt; breadth first search
            [see] =&gt; 
            [description] =&gt; 
        )

    [breadth first search] =&gt; Array
        (
            [slug] =&gt; breadth first search
            [parent] =&gt; 
            [name] =&gt; breadth first search
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; In breadth first search of a {{search tree}}, one prgresses by first of all looking at all of the nodes at level 1 (one step form the root), then all the nodes at level 2, etc. 
        )

    [byzantine conditions] =&gt; Array
        (
            [slug] =&gt; byzantine conditions
            [parent] =&gt; 
            [name] =&gt; Byzantine conditions
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; Byzantine conditions are apparently rare (but in practice possibly not) confluences of parameters, variables or situations which give rise to unexpectedly bad consequences.  For example, some {{sortling algorithms}} behave particularly badly when given an iniitally fully sorted list of values.  Most algorithms have worst case scenarios, when they behave very badly, but so ling as they are rare this is accpeted.  However, soemtimes the structure of processes that give rise to input data means these rare situations happen surprisingly ooftem -- the case of a sorted list is one.  &lt;br&gt;&lt;br&gt; One way to avoid Byzantine conditions is to insert some form of {{randomness}} or {[stochastic}} choices.  For example, a {[dterministic}} version of {{quicksort}} uses the first element as a separator to divide the rest of the list into those greater and smaller.  This leads precisely to a worst case situation with a sorted list as input -- a Byzantine condition.  If instead the separator is chosen as a {{random}} item from the list, the worst case can still happen, but is vanishingly rare.
        )

    [c4.5] =&gt; Array
        (
            [slug] =&gt; c4.5
            [parent] =&gt; 
            [name] =&gt; C4.5
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; C4.5 is one of the most well known variants of {{Quinlan}}'s {{ID3}} {{decision tree}} learning algorithm.  C4.5 has a number of variations from basic ID3 including the use of {{normalised information gain}} for non-binary node decisions.
        )

    [cambridge analytica scandal] =&gt; Array
        (
            [slug] =&gt; cambridge analytica scandal
            [parent] =&gt; 
            [name] =&gt; Cambridge Analytica scandal
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; Cambridge Analytica was a counsulting company that infamously used large amounts of data gathered from Facebook to help target political advertising.  The extent of this was uncovered in 2018 including its use in the 2016 Trump presidential campaign.  The use was deemed to have violated {{privacy}} and various {{data protection laws}} and Facebook was subsequently fined in both the UK and USA.  
        )

    [cartesian coordinates] =&gt; Array
        (
            [slug] =&gt; cartesian coordinates
            [parent] =&gt; 
            [name] =&gt; Cartesian coordinates
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; Cartesian coordinates are the rectilinear x,y (2D) or x,y,z (3D) cooridinates we see in graphs and maps.  These are a common way to represent locations on a plane or in space, or pixels in an image, but sometimes alternatives are used, for example, ((radial coordinates}} where (in 2D) each point. is represented as a distance and angle from centre point.  Soemtimes a {{change in representation}} can make mathematical analysis or {{machine learning}} easier.
        )

    [case grammar] =&gt; Array
        (
            [slug] =&gt; case grammar
            [parent] =&gt; 
            [name] =&gt; case grammar
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; 
        )

    [case grammars] =&gt; Array
        (
            [slug] =&gt; case grammars
            [parent] =&gt; 
            [name] =&gt; case grammars
            [canonical] =&gt; case grammar
            [see] =&gt; 
            [description] =&gt; 
        )

    [case-based reasoning] =&gt; Array
        (
            [slug] =&gt; case-based reasoning
            [parent] =&gt; 
            [name] =&gt; case-based reasoning
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; Case-based reasing uses a {{database}} of examples (cases) andwhen it encounters a new situation matces it to the database, retreives the closest past examples, and attempts to understand the current situation in the light of these past cases.  Doing this usually involves identifying the differences between the currrent situation and retreived case and working out the imoacts of this using some other form of {{reasoning}}.
        )

    [cbr] =&gt; Array
        (
            [slug] =&gt; cbr
            [parent] =&gt; 
            [name] =&gt; CBR
            [canonical] =&gt; case-based reasoning
            [see] =&gt; 
            [description] =&gt; 
        )

    [cellular automata] =&gt; Array
        (
            [slug] =&gt; cellular automata
            [parent] =&gt; 
            [name] =&gt; cellular automata
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; Cellular automata are simple computational elements laid out in a spatial pattern, most commonly a square grid, where each automata's state updates based only on its own state and that of its neighbours.  The most well known example is {{Comway}}'s {{Game of Life}}. Very simple rules can give rise to complex {{emergent behaviour}}.
        )

    [central example] =&gt; Array
        (
            [slug] =&gt; central example
            [parent] =&gt; 
            [name] =&gt; central example
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; When. looking at a class or cluster, a central example is one that is far form the edges, so that small {{pertubations}} in the values in the example do not change the classification.  For example, a blackbird is a &lt;em&gt;central example&lt;/em&gt; of a bird, wereas a penguin (lfughtless swims a lot) owuld be a {{boundary example}}.  Cenral exmaples can be useful in {{explainable AI}} when the rules for allocating classes have been generaed using {{machine learning}}.
        )

    [centroid] =&gt; Array
        (
            [slug] =&gt; centroid
            [parent] =&gt; 
            [name] =&gt; centroid
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; The centrid is the cdentre of a group of points as cacuulated using the {{arithmetic mean}} of each of the {{Cartesian coordinates}} of the points.  For example, in 2D sace, if the points have coordinates [ (2,0), (2,2), (3,1), (1,1) ] then the centroid is ( 2, 1 ), because 2 is the mean of the x coordinates: 2, 2, 3 and 1 and 1 is the mean of the y coordinates: 0, 2, 1, 1.
        )

    [cerebral cortex] =&gt; Array
        (
            [slug] =&gt; cerebral cortex
            [parent] =&gt; 
            [name] =&gt; cerebral cortex
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; The cerebral cortex is the deeply folded outer layer of the brain (also known as {{white matter}}).  It is a large sheet of neurons in six rough layers, with different kinds of neurons in each layer
        )

    [cern] =&gt; Array
        (
            [slug] =&gt; cern
            [parent] =&gt; 
            [name] =&gt; CERN
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; CERN is the international particle accelerator and research centre in Switzerland where the {{World Wide Web}} began.

        )

    [certainty factors] =&gt; Array
        (
            [slug] =&gt; certainty factors
            [parent] =&gt; 
            [name] =&gt; certainty factors
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; Certainty factors are way of performing {{uncertain reasoning}}.  While {{Bayesian reasinong}} demands that belief be precisely coded as a single {{probability}}-like number, confidence factors allow a range with belief for and against a particlar conclusion to build as new evidence is found.   Initially , each initial fact/evidence is initially given a certainty factor where positve in the range +1  to -1, where positive values denote evdience for and negative evidence againts, worth zero meaning no evidence and the extremes +1/-1 meaning incontrovertable evidence for/against).  Derived facts are given two values, a measure of belief and a measure of diselief (each [0,1]) based on evidence, and these are accumulated as further positive or negative evidence arises.   This works well in medical domains where there can be evidence both for and against a diagnosis.    Certainty factors were used in the successful early {{expert system}} {{MYCIN} and still used in {{expert systems}} including {{neural network}} variants.  However, they are less popular now than {{Bayesian methods}} and {{fuzzy reasoning}}.
        )

    [changes in status] =&gt; Array
        (
            [slug] =&gt; changes in status
            [parent] =&gt; 
            [name] =&gt; changes in status
            [canonical] =&gt; 
            [see] =&gt; status change event
            [description] =&gt; 
        )

    [charles babbage] =&gt; Array
        (
            [slug] =&gt; charles babbage
            [parent] =&gt; 
            [name] =&gt; charles babbage
            [canonical] =&gt; babbage-charles
            [see] =&gt; 
            [description] =&gt; 
        )

    [charniak] =&gt; Array
        (
            [slug] =&gt; charniak
            [parent] =&gt; 
            [name] =&gt; charniak
            [canonical] =&gt; charniak, eugene
            [see] =&gt; 
            [description] =&gt; 
        )

    [charniak, eugene] =&gt; Array
        (
            [slug] =&gt; charniak, eugene
            [parent] =&gt; 
            [name] =&gt; Eugene Charniak
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; Eugene Charniak was a Professor at Brown University and an early AI pioneer.  Aminsgt other thins he worked on the way common sense knowlegde is needed to interpret simple chuldren's stories.
        )

    [chatbots] =&gt; Array
        (
            [slug] =&gt; chatbots
            [parent] =&gt; 
            [name] =&gt; chatbots
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; Chatbots are a form of software that converses using text-based communication in a (more or less) human-like manner.  The earliest chatbots such as {{ELIZA}} used very simple pattern matching, and more compelx variants of this are widely used for customer support on websites.  Chatbots, such as {{ChatGPT}} powered by {{large language modles} have transformed the fluidity of chatbots.   Speech-based chatbots are also possible, but are more often known by different terms, such as {{conversational agents}}.
        )

    [chatgpt] =&gt; Array
        (
            [slug] =&gt; chatgpt
            [parent] =&gt; 
            [name] =&gt; ChatGPT
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; ChatGPT is a {{chatbot}} based on {{OpenAI}}'s {{large language models}} {{GPT-3}} and {{GPT-4}}.  Early experiences with ChatGPT transformed tpublic and media perc eptions of AI as it is able to produce text that is not only {grammatically correct, but also appears to understand concepts.  That is, it arguably passes the {{Turing test}} as it can be indistinguishable from human text-based chat.  In additon, the ability to create good-enough student essays, news stories and even film-scripts has caused considerable concern about academic integrity, accuracy of news, and job security of creative professionals.
        )

    [chess] =&gt; Array
        (
            [slug] =&gt; chess
            [parent] =&gt; 
            [name] =&gt; chess
            [canonical] =&gt; 
            [see] =&gt; computer chess
            [description] =&gt; 
        )

    [childhood cognitive development] =&gt; Array
        (
            [slug] =&gt; childhood cognitive development
            [parent] =&gt; 
            [name] =&gt; childhood cognitive development
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; Childhood cognitive development describes the way that small children's ability to think, reason and learn develops from birth to adulthood, in particular the early years.  Many studies of models of {{human reasoning}} focus on adults, but there are often substantial differences of kind not just level in childhood.  Each person develops differently and at a different pace, but general patterns have been studied.  In particular, {{Piaget}}'s work, which was based on studies of his own children and those in an oprphanage, proposed a linear developent of different cignitive abilities, that proceeds at different paces, but is the same for all children.  While {{Piaget}}'s work is still seminal, the details are challenged and there are now more nuanced views of development.
        )

    [chinese room] =&gt; Array
        (
            [slug] =&gt; chinese room
            [parent] =&gt; 
            [name] =&gt; Chinese room
            [canonical] =&gt; chinese room argument
            [see] =&gt; 
            [description] =&gt; 
        )

    [chinese room argument] =&gt; Array
        (
            [slug] =&gt; chinese room argument
            [parent] =&gt; 
            [name] =&gt; Chinese room argument
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; The Chinese room is a {{thought experiment}} created by the philosopher {{John Searle}} to argue that true machine intelligence is impossible,  In partcular it seeks  to counter the {{physical symbol hypothesis}}, and deminstrate that intelligent behaviour is not the same as {{intelligence}.  Searle posits a sealed room wthin which a person who has no understanding of the Chinese languge manipulates tables of symbols using written rules.  The rules are such that whenever a question in Chinese is posted into the room, the person is able to use the rules to create a meaningful response in Chinese to be posted back out.  Those outside the room might believe that someone in the room understands Chienses because the room as whole behaves like this.  However, there is no {{intentionality}}, no true meaning.  The person can perform manipulations that appear to understand Chinese, but there is no actual understanding.  Counters to this are based laregly on the {{system response}}, that the room as a whole can 'understand' even of the person does not.
        )

    [chunking] =&gt; Array
        (
            [slug] =&gt; chunking
            [parent] =&gt; 
            [name] =&gt; chunking
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; Chunking is used in {{cognitive science}} to refer to the way that groups of items can be tretaed as a single unit (a chunk).  This occurs in {{short-term memory}}, where it is a lot eaiser to remember 5 words than 30 random letters.   Chunking is also used to refer to the active process whereby frequent groups of stiumuli, patterns of thought or well practiced movements become treated as a single chunk.  This can be emulated in computational {{cognitive architectures}} including in {{SOAR}}.
        )

    [clarke, arthur c] =&gt; Array
        (
            [slug] =&gt; clarke, arthur c
            [parent] =&gt; 
            [name] =&gt; Clarke, Arthur C.
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; Arthur C. Clarke is a science fiction author, who wrote '2001 a Space Odyssey' (novel and film) that includes the {{sentient}} compter {{HAL}}.
        )

    [clippy] =&gt; Array
        (
            [slug] =&gt; clippy
            [parent] =&gt; 
            [name] =&gt; Clippy
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; Clippy was an {{intelligent agent}} that was deployed for a period in versions of Microsoft Office (officially called Cliipit).  It was depicted using an icon with speach bubbles, that looked like a paper clip with a face.  Clippy would monitor your work and make useful suggestions.  However, the design of the {{user interface}} for Clippy meant it would interupt your work, so very annoying when the suggestion was not needed.  That is it did NOT embody {{appropriate intelligence}}.
        )

    [closed list] =&gt; Array
        (
            [slug] =&gt; closed list
            [parent] =&gt; 
            [name] =&gt; closed list
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; In {{search algorithms}} the closed list records the set nodes that have already been considered.  It is required in {{graph search}} to avoid re-exmaining the same nodes.  This is in contrast to the {{open-list}} which records nodes that still need to be explored.
        )

    [closed world assumption] =&gt; Array
        (
            [slug] =&gt; closed world assumption
            [parent] =&gt; 
            [name] =&gt; closed world assumption
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; When a {{reasoning system}} adopts the closed world assumption, it assumes that all facts (within its domain) are present in its {{knowledge base}}.  That is if a fact is not in the {{knowledge base}}, it is deemed to be false.  The {{logic programming}} language {{Prolog}} adopts a closed world assumption.
        )

    [closed-loop control] =&gt; Array
        (
            [slug] =&gt; closed-loop control
            [parent] =&gt; 
            [name] =&gt; closed-loop control
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; Closed-loop control is a term used in {{cybernetics}} and {{control engineering}} to describe a system that monitors the impact of its outpits and uses this to adjust its behaviour.  A simple exmaple is the Centrifugal governor in steam engines.  As the engine move faster it spins a mechanism with two heavy balls attached. As they spin they rise up opening.a vavlue that reduce the pressure and hence maintains the speed close to a desired level.
        )

    [cloud computation] =&gt; Array
        (
            [slug] =&gt; cloud computation
            [parent] =&gt; 
            [name] =&gt; cloud computation
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; Cloud computing is when computation is performed in remote {{data centres}} rather than in known, local computers.  In some cases you have one or more dedicated computers, on some computers (or virtual computers) are started and stopped on request, and sometimes code is completely serverless, executing wherever ther is capacity.  In most cases the physical infrastructure is managed by a separate company who provide {{compute as a service}}.
        )

    [clustering] =&gt; Array
        (
            [slug] =&gt; clustering
            [parent] =&gt; 
            [name] =&gt; clustering
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; Clustering algorithms take data and tries to sort it into clusters each with some form of similar characteristics.  This is like you might do if looking a scatter graph and noticing that the data falls into two or three groups.  Examples of cluster analysis includes {{k-means}} which is based around calculating centroids of each cluster and {{Khonne networks}} (or {{aelf-organsing maps}}) which are a form of neural netwrok where similar data points are classified close to one another a two dimensional grid.  Clustering algorithms are a form of {{unsupevsied learnong}}, but there are also {{semi-supervised}} forms where some or all of the training set has a classification and the clustering algorithm tries to ensure that each cluster has a consistent label, even if there may be several clusters for each label.
        )

    [clusters] =&gt; Array
        (
            [slug] =&gt; clusters
            [parent] =&gt; 
            [name] =&gt; clusters
            [canonical] =&gt; clustering
            [see] =&gt; 
            [description] =&gt; 
        )

    [co-operating agents] =&gt; Array
        (
            [slug] =&gt; co-operating agents
            [parent] =&gt; 
            [name] =&gt; co-operating agents
            [canonical] =&gt; 
            [see] =&gt; cooperating agents
            [description] =&gt; 
        )

    [cognitive architecture] =&gt; Array
        (
            [slug] =&gt; cognitive architecture
            [parent] =&gt; 
            [name] =&gt; cognitive architecture
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; A  cognitive architecture is a computational model that attempts to emulate real human cognition, but at the level of larger unuts than {[artificial neurla networks}}.  For example, a cognitive architecture may have a single (complex) unit to represent {{working memory}}.  This can be used as a awy to understand {{human cognition}} or as inspiration for artificial human-like intelligence.   Influential example include {{SOAR}} and {{ACT*}} {{ACT-R}}.  The {[regret model}} described in the book also has a form of cognitive architure including units for emotional response and counter-factual reasoning.
        )

    [coin-weighing problem] =&gt; Array
        (
            [slug] =&gt; coin-weighing problem
            [parent] =&gt; 
            [name] =&gt; coin-weighing problem
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; Coin-weighing problems and similar puzzles are often used as simple examples for education or as tests for algorithms.  Typical weighing tasks have some number of coins (or other ietms) that should all be identical, but where one (or more) is known to be heavier, lighter or simply diffeernt from the others (in the case of coins usually a counterfeit).  The goal is usually to find the counterfeit(s) using as few weighings as possible.
        )

    [combinatorial explosion] =&gt; Array
        (
            [slug] =&gt; combinatorial explosion
            [parent] =&gt; 
            [name] =&gt; combinatorial explosion
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; Combinatorial explosion is the term applied to certain kinds of problem where the complexity grows very rapidly.  This arises when all combinations of, even relatively small, sets of elements need to be considered.  For example, imagine we are looking at choices of flavours in ice-cream sundaes (ignoring how much, it is a very generous gelateria).  If we have two flavours, strawberry and banana, then there are only four options: plain strawberry; plain banana; strawberry and banana; and (for the restrained) none at all.  If there are three flavours, there are 8 options; for four flavours 16; and for a gelateria with a dozen flavours more than 4096.  
        )

    [common ground] =&gt; Array
        (
            [slug] =&gt; common ground
            [parent] =&gt; 
            [name] =&gt; common ground
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; Common ground is a term, coined by Herbert (Herb) Clark and colleagues, used in the theory of cooperative communications.  It refers to the way that two or more people establish a level of shared understanding sufficient for theor purposes as a side effect of communicating to acheive theor goals.  They typically do not try to have complete shared knowledge of everything, nor even spend lots of time at the beginning establishing shared language, but instead build this through unconscious but skillful use of language includuing {{redundancy}} and {{repair}}.  This can be used as a model for human-AI interactions and inter-agent communications.
        )

    [compas] =&gt; Array
        (
            [slug] =&gt; compas
            [parent] =&gt; 
            [name] =&gt; COMPAS
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; COMPAS is an AI based system to predict recidivism, that is the likelihood that someone reoffends after leaving prison.  It was traied on US data and has been used, controversially, in US courts as part of parole decsion making.  Analysis has suggested that the outputs may embody a degree of racial {{bias}}.
        )

    [compliant motion] =&gt; Array
        (
            [slug] =&gt; compliant motion
            [parent] =&gt; 
            [name] =&gt; compliant motion
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; 
        )

    [computer chess] =&gt; Array
        (
            [slug] =&gt; computer chess
            [parent] =&gt; 
            [name] =&gt; computer chess
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; For many years computer chess was one of the grand challenges in AI, until, in 1996 the {{IBM}} chess computer {{Big Blue}} defeated the chess grandmaster Garry Kasparov.
        )

    [concept learning] =&gt; Array
        (
            [slug] =&gt; concept learning
            [parent] =&gt; 
            [name] =&gt; concept learning
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; Concept learning is when the coal of learing is to allocate input into one of a number of distinct classifications. Often concept learing is applied to. binary condition 'in category' vs 'not in category', and some algorithms or typically expressed in this binary form.  Examples of concept learning include symbolic algirthms such as {{version spaces}} and  {{ID3}}, and also sub-symbolic technqies including many kinds of {{neural network}} and {{swarm computing}}.
        )

    [confidence] =&gt; Array
        (
            [slug] =&gt; confidence
            [parent] =&gt; 
            [name] =&gt; confidence
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; 
        )

    [configuration parameters] =&gt; Array
        (
            [slug] =&gt; configuration parameters
            [parent] =&gt; 
            [name] =&gt; configuration parameters
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; 
        )

    [configuration space] =&gt; Array
        (
            [slug] =&gt; configuration space
            [parent] =&gt; 
            [name] =&gt; configuration space
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; 
        )

    [confusion matrix] =&gt; Array
        (
            [slug] =&gt; confusion matrix
            [parent] =&gt; 
            [name] =&gt; confusion matrix
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt;  Confusion matrix is a {{square matrix}} used to assess the accuracy of {{classification}} methods.  The rows are labelled by the class given by the {{classifier}} and the columns are labelled by the actual class.   Each cell is the {{probability}} (or sometimes {{percentage}} or {{freqeuncy}}) with with which an example that is classified as a partcular class is actually of another.  The diagonal represents correct classifications and the off-diagonal elements errors.  The numbers on a row shoudl add up to 1 (for probabilities) or 100% (for percentages).  In order to turn this information into actual accuracy fifgures, one needs to know about the {{base rate}} for each class.  For example, the diagonal entries could be largely near 100% (highly accureate) for all but one case, but that case might be the most common in practice.
        )

    [connectionist] =&gt; Array
        (
            [slug] =&gt; connectionist
            [parent] =&gt; 
            [name] =&gt; connectionist
            [canonical] =&gt; connectionist model
            [see] =&gt; 
            [description] =&gt; 
        )

    [connectionist model] =&gt; Array
        (
            [slug] =&gt; connectionist model
            [parent] =&gt; 
            [name] =&gt; connectionist model
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; 
        )

    [connectionist model of cognition] =&gt; Array
        (
            [slug] =&gt; connectionist model of cognition
            [parent] =&gt; 
            [name] =&gt; connectionist model of cognition
            [canonical] =&gt; connectionist model
            [see] =&gt; 
            [description] =&gt; 
        )

    [connectionist models] =&gt; Array
        (
            [slug] =&gt; connectionist models
            [parent] =&gt; 
            [name] =&gt; connectionist models
            [canonical] =&gt; connectionist model
            [see] =&gt; 
            [description] =&gt; 
        )

    [connectionist system] =&gt; Array
        (
            [slug] =&gt; connectionist system
            [parent] =&gt; 
            [name] =&gt; connectionist system
            [canonical] =&gt; connectionist model
            [see] =&gt; 
            [description] =&gt; 
        )

    [connectionist systems] =&gt; Array
        (
            [slug] =&gt; connectionist systems
            [parent] =&gt; 
            [name] =&gt; connectionist systems
            [canonical] =&gt; connectionist model
            [see] =&gt; 
            [description] =&gt; 
        )

    [connectionist system!as model of cognition] =&gt; Array
        (
            [slug] =&gt; connectionist system!as model of cognition
            [parent] =&gt; connectionist system
            [name] =&gt; as model of cognition
            [canonical] =&gt; connectionist model
            [see] =&gt; 
            [description] =&gt; 
        )

    [consequent-driven reasoning] =&gt; Array
        (
            [slug] =&gt; consequent-driven reasoning
            [parent] =&gt; 
            [name] =&gt; consequent-driven reasoning
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; 
        )

    [conservative warnings] =&gt; Array
        (
            [slug] =&gt; conservative warnings
            [parent] =&gt; 
            [name] =&gt; conservative warnings
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; 
        )

    [constraint propagation] =&gt; Array
        (
            [slug] =&gt; constraint propagation
            [parent] =&gt; 
            [name] =&gt; constraint propagation
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; 
        )

    [constraint satisfaction] =&gt; Array
        (
            [slug] =&gt; constraint satisfaction
            [parent] =&gt; 
            [name] =&gt; constraint satisfaction
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; 
        )

    [constraints] =&gt; Array
        (
            [slug] =&gt; constraints
            [parent] =&gt; 
            [name] =&gt; constraints
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; 
        )

    [context] =&gt; Array
        (
            [slug] =&gt; context
            [parent] =&gt; 
            [name] =&gt; context
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; 
        )

    [context-aware interaction] =&gt; Array
        (
            [slug] =&gt; context-aware interaction
            [parent] =&gt; 
            [name] =&gt; context-aware interaction
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; 
        )

    [context-free grammar] =&gt; Array
        (
            [slug] =&gt; context-free grammar
            [parent] =&gt; 
            [name] =&gt; context-free grammar
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; A contewxt free grammar is described using hierarchical rules.  They are used extensively for formal notations such as computer programming languages.  They can also be applied to {{natural language processing}}, but have limitations   The hierarchical rules show how a {{non-terminal}} term expands into (or is composed of) some combination of {{terminals}} other {{non-terminals}}.  For example, part of the expansion of a Noun_Phrase ({{non-terminal}}) is:  &lt;br&gt; &amp;nbsp; &amp;nbsp; &amp;nbsp; Noun_Phrase &amp;nbsp; &amp;nbsp; := &amp;nbsp; &amp;nbsp; &quot;the&quot; &amp;nbsp; Qualified_Noun&lt;br&gt; On the right 'the' is a {{terminal}}, a single word and the end point of grammtical expansion, whereas Qualified_Noun is a {{non-terminal}}, which could end up being a single word (e.g. &quot;ball&quot;), but also has expansions including possessives, adjectives, etc. (e.g. &quot;Bella's blue ball&quot;).  Note that the rules apply no matter in what context the rule is used, hence 'context free'.  However, whether a rule is applied in the first place does depend on previosuly applied grammatical rules.  For example one would not attempt to apply the rules for Verb after &quot;the&quot;.  
        )

    [context-free grammars] =&gt; Array
        (
            [slug] =&gt; context-free grammars
            [parent] =&gt; 
            [name] =&gt; context-free grammars
            [canonical] =&gt; context-free grammar
            [see] =&gt; 
            [description] =&gt; A contewxt free grammar is described using hierarchical rules.  They are used extensively for formal notations such as computer programming languages.  They can also be applied to {{natural language processing}}, but have limitations   The hierarchical rules show how a {{non-terminal}} term expands into (or is composed of) some combination of {{terminals}} other {{non-terminals}}.  For example, part of the expansion of a Noun_Phrase ({{non-terminal}}) is:  &lt;br&gt; &amp;nbsp; &amp;nbsp; &amp;nbsp; Noun_Phrase &amp;nbsp; &amp;nbsp; := &amp;nbsp; &amp;nbsp; &quot;the&quot; &amp;nbsp; Qualified_Noun&lt;br&gt; On the right 'the' is a {{terminal}}, a single word and the end point of grammtical expansion, whereas Qualified_Noun is a {{non-terminal}}, which could end up being a single word (e.g. &quot;ball&quot;), but also has expansions including possessives, adjectives, etc. (e.g. &quot;Bella's blue ball&quot;).  Note that the rules apply no matter in what context the rule is used, hence 'context free'.  However, whether a rule is applied in the first place does depend on previosuly applied grammatical rules.  For example one would not attempt to apply the rules for Verb after &quot;the&quot;.  
        )

    [context-sensitive grammar] =&gt; Array
        (
            [slug] =&gt; context-sensitive grammar
            [parent] =&gt; 
            [name] =&gt; context-sensitive grammar
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; 
        )

    [context-sensitive grammars] =&gt; Array
        (
            [slug] =&gt; context-sensitive grammars
            [parent] =&gt; 
            [name] =&gt; context-sensitive grammars
            [canonical] =&gt; context-sensitive grammar
            [see] =&gt; 
            [description] =&gt; 
        )

    [continuous distribution] =&gt; Array
        (
            [slug] =&gt; continuous distribution
            [parent] =&gt; 
            [name] =&gt; continuous distribution
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; 
        )

    [contour following] =&gt; Array
        (
            [slug] =&gt; contour following
            [parent] =&gt; 
            [name] =&gt; contour following
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; 
        )

    [convariance] =&gt; Array
        (
            [slug] =&gt; convariance
            [parent] =&gt; 
            [name] =&gt; convariance
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; 
        )

    [converge] =&gt; Array
        (
            [slug] =&gt; converge
            [parent] =&gt; 
            [name] =&gt; converge
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; An {{iterative algorithm}} is said to converge if its results gradually vary less and become closer and closer to some final value.  Not that the former does not imply the latter, for example in the sum 1+1/2+1/3+1/4... the partial answers get closer and closer to each, but the sum does not  converge to a final value.  In contrast the sum 1+1/2+1/4+1/8... does converge as the partial sums values get closer and closer (but never reach) to the value 2.
        )

    [convolutional neural network] =&gt; Array
        (
            [slug] =&gt; convolutional neural network
            [parent] =&gt; 
            [name] =&gt; convolutional neural network
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; 
        )

    [convolutional neural networks] =&gt; Array
        (
            [slug] =&gt; convolutional neural networks
            [parent] =&gt; 
            [name] =&gt; convolutional neural networks
            [canonical] =&gt; convolutional neural network
            [see] =&gt; 
            [description] =&gt; 
        )

    [convolutions] =&gt; Array
        (
            [slug] =&gt; convolutions
            [parent] =&gt; 
            [name] =&gt; convolutions
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; 
        )

    [conway] =&gt; Array
        (
            [slug] =&gt; conway
            [parent] =&gt; 
            [name] =&gt; conway
            [canonical] =&gt; conway, john
            [see] =&gt; 
            [description] =&gt; 
        )

    [conway, john] =&gt; Array
        (
            [slug] =&gt; conway, john
            [parent] =&gt; 
            [name] =&gt; Conway, John
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; John Conway was a mathematician, who, amongst other things, created the {{Game of Life}}, a simple {{cellular automata}} with surprisingly rich behaviour.
        )

    [conway's game of life] =&gt; Array
        (
            [slug] =&gt; conway's game of life
            [parent] =&gt; 
            [name] =&gt; Conway's Game of Life
            [canonical] =&gt; game of life
            [see] =&gt; 
            [description] =&gt; 
        )

    [cooperating agents] =&gt; Array
        (
            [slug] =&gt; cooperating agents
            [parent] =&gt; 
            [name] =&gt; cooperating agents
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; Software {{agents}} may operate entirely independently performing discrete tasks.  However, often multiple agants cooperate on a single task, or share useful infornation with one another.  This can be done using some form of coordinating agent, but more often more loosely coupled.  Communications between agents may be direct using {{messages}} or invoking {{methods}}, or indirect through a shared communication means such as a {{blackboard architecture}}, allowing {{emergent behaviour}} beyond that coded into any individual agent.  The agents may execute within a single machine or in {{distributed AI}} over multiple machines, maybe across the whole internet.  &lt;br&gt;&lt;br&gt;. Multiple agents operating in an environment may affect one another indirectly, and so give rise to unexpected {{emergent behaviour}}, but the term &lt;em&gt;cooperating agents&lt;/em&gt; would normally be reserved for situations where there is some form of explicit or planned cooperation.
        )

    [correlation matrix] =&gt; Array
        (
            [slug] =&gt; correlation matrix
            [parent] =&gt; 
            [name] =&gt; correlation matrix
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; A correlation matrix is fored when you have multiple {{features}} on a stream or set of data, for example multiple {{database}} records or {{time series}}.  The {{correlation cooefficient}} is calculated for each pair of features and set in a matrix.  The matrix is symmetric as the  {{correlation cooefficient}} does not depnd on the order of the features, and the diagonal is the correlation between a feature and itself.  &lt;br&gt;&lt;br&gt; The correlation matrix can be used as a means to visualise the rekationships between featyres, and in oarticukar if there are some goruos of features that are all closely related with one anther.  It can also be used as a {{pre-processing}} step before other analysis such as {{clustering}} of features or {{dimensional reduction}}.
        )

    [cosine similarity] =&gt; Array
        (
            [slug] =&gt; cosine similarity
            [parent] =&gt; 
            [name] =&gt; cosine similarity
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; The cosine similarity measures how close two {{vectors}} are to one another using the cosine of the angle between them.  Unlike a {{Euclidean distance}}, the cosine similarity ignores the size of the two vectors, it just focuses on their direction.  In {{document retrieval}] and {{text processing}}, the cosine silarity is applied to the vector representing the frequencies of words in a document, soemtimes weighted by overalll {{corpus frequency}}.
        )

    [cost-benefit] =&gt; Array
        (
            [slug] =&gt; cost-benefit
            [parent] =&gt; 
            [name] =&gt; cost&amp;ndash;benefit
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; There is often a cost--benefit balance or cost--benefit {{trade-off}} as decsions that lead to benefots may also have associated costs.  We have to chose whether the benefit is worth the cost.  This is easy if both are certain, in the same units and at the same time.  For example, suppose choice A has a benfit of 100 euro and cost of 50 euro and choice B has a benefit of  70 euro and cost of 10 euro; it is clear that choice A is better.  However, the benefit may depend on the wetaher or the throw of a dice, so that weighing up cost/benefot may involve {{risk}}. It gets more cmplicated when the units of value differ for exmaple if the  benefit is be monetary and the cost about health and well being, such as choosing between a well-paid but boring and stressful job versus an fulfilling job that pays very little.    The costs and benefots might also arose at different times, for example, saving in a pension fund.
        )

    [counterfactual reasoning] =&gt; Array
        (
            [slug] =&gt; counterfactual reasoning
            [parent] =&gt; 
            [name] =&gt; counterfactual reasoning
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; Counterfactual reasoning is when you reason about something that you believe to be false, often about different possible ways the world could be or culd have been.  This might be about past events, &quot;if it hadn't rained it would have been a good day&quot;, or about enitirely fiictional sitiuations &quot;if mice could fly ...&quot;.  In classical {{logic}} this leads to contradicitons, unless one is careful to ensure that the true contradictory facts are not part of our reasoning process.  This is used in mathematics in {{proof by contradiction}}.  In day-to-day counterfactual reasoning, we are more able to compartmentalise, so that in a story about &quot;if mice could fly&quot;, we would avoid thinsg that depended on the (true) fact that mice don't fly.   In {{regret}} we use counterfactual reasoning to weight up the potential advanategs or costs if we had chosen alternative actions &quot;if only I hadn't walked under the ladder ...&quot;.
        )

    [cousin relationship] =&gt; Array
        (
            [slug] =&gt; cousin relationship
            [parent] =&gt; 
            [name] =&gt; cousin relationship
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; The meaning of 'cousin' varies between cultures, but within the {{AI book}} we adopt the definition that X is a cousin of Y if X is the offsping of a sibling of the parent of Y.  For example the daughter of the brother of your mother is a cousin.  In {{Prolog}} we might express this as:  &lt;br&gt; &amp;nbsp; &amp;nbsp; &amp;nbsp; cousin{X,Y)  :-  child(X,A), sibling(A,B), child(Y,B). &lt;br&gt;  Note that one has to be careful in translating these definitions into formal logics, as aspects that are 'obvious' the human may not be to a machine.  For example, if (say in {{Prolog}}), we have parent child relationships, we might think that, &quot;X is a sibling of Y&quot; could be expressed  as: &lt;br&gt; &amp;nbsp; &amp;nbsp; &amp;nbsp; sibling(A,B)  :-  child(A,P), child(B,P). &lt;br&gt;  That is, a sibling of the child of your parent. However, this would mean that  you are your own sibling (see code example tbirds.p), and hence when we put this into the cousin relationship, {{Prolog}} would conclude that your sister is a cousin or even that you are a cousin of yourself.   We think &quot;but of course when you say your sibling is a chld of your parent, you don't mean yourself!&quot;, but this has to be explicitly stated in a formal system. 
        )

    [credit assignment] =&gt; Array
        (
            [slug] =&gt; credit assignment
            [parent] =&gt; 
            [name] =&gt; credit assignment
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; Credit assignment in AI is about working out which which rule or parameter should be given the credit (or blame)  when something works well or badly. This is important to enable the syetm to learn, by reinforcing good rules.  However, this can be hard when several rules are involved, hence one of the reaosn for {{explanation-based learning}}.  It can be especially difficult in {{reinformcement learning}} as the reward may occur sometime after the crirtial action with other actiosn in between.  For example, if you take a wrong turn early in a journey it may be some time before you realise, but the last few navigation chocies may nave been reasonable &ndash; it is the first choice that should be given the 'credit' ( this case approbation).
        )

    [credit assignment problem] =&gt; Array
        (
            [slug] =&gt; credit assignment problem
            [parent] =&gt; 
            [name] =&gt; credit assignment problem
            [canonical] =&gt; credit assignment
            [see] =&gt; 
            [description] =&gt; 
        )

    [critical point] =&gt; Array
        (
            [slug] =&gt; critical point
            [parent] =&gt; 
            [name] =&gt; critical point
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; The term critical point is used in roughly its day-to-day meaning in a variety of contexts, usually where there is some qualitative change in behaviour, such as a phase change with temeprature (e.g. solid/liquid) or crystalisation.  When {{nanoparticle}} deposition is used in {{reservoir computing}} there is a critical point as the susbtrate changes from having no conduction paths to total conduction; at that point it has the most interesting and useful {{quantum}} behaviours.
        )

    [cross-validation] =&gt; Array
        (
            [slug] =&gt; cross-validation
            [parent] =&gt; 
            [name] =&gt; cross-validation
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; Cross-validation is when you retain some portion of a training set to verify learning on the rest.  It is a way to test the {{generalisation}} of the resulting {{machine learning}} or {{statistical model}}.  Typically models perform better on the data on which they are trained than on fresh data, but can {{overfit}} the data.  When the goal is to evaluate an algorithm {{k-fold cross validation}} may be used where the data is divided into k subsets (folds) and the cross-validation is perfoemd k tiemes, in each pass withholding out one of the folds from the rtainung set and using ot to validate the outcome..
        )

    [crowdsourcing] =&gt; Array
        (
            [slug] =&gt; crowdsourcing
            [parent] =&gt; 
            [name] =&gt; crowdsourcing
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; Crowdsourcing is when the general public or other large group collaboratively create a large data set.  This may be done as an exlicit goal of the project, for example Wikipedia editors, or BBC Spring Watch, which asks viewers to post their forst sightings of wildlife in the spring.  It can also occur implicitly as part of other activities, for example in the iSpot nature application people post photos or descriptions of wildlife sightings which other people answre; over time this collectively builds a large knowledge base for the users to consult.  Crowdsourcing is related to {{human computation}}, but the connotations of the latter treat people more like cogs in a machine whereas crowdsourcing usually suggest a level of personal interest..
        )

    [cuban missile crisis] =&gt; Array
        (
            [slug] =&gt; cuban missile crisis
            [parent] =&gt; 
            [name] =&gt; Cuban missile crisis
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; The Cuban missile crisis was a stand-off in 1962 between the USA and USSR over the sitting of Russian nuclear missiles in Cuba, which would place them in close range to mainland USA, like the US ones in Europe were to Russia.  The world got close to nucelaer war, an outcome that was worst for everyone, but for each side they could lose if they backed down first.  It can be seen as a real-life version of te {prisoners' dilemma}}, excpet that happily the US and USSR leadership could talk to one another and find a way to resolve the conflict.  For AI illustrates the importance of {{agent communication}}.
        )

    [current state] =&gt; Array
        (
            [slug] =&gt; current state
            [parent] =&gt; 
            [name] =&gt; current state
            [canonical] =&gt; current state representation
            [see] =&gt; 
            [description] =&gt; 
        )

    [current state!representation] =&gt; Array
        (
            [slug] =&gt; current state!representation
            [parent] =&gt; current state
            [name] =&gt; representation
            [canonical] =&gt; current state representation
            [see] =&gt; 
            [description] =&gt; 
        )

    [current state representation] =&gt; Array
        (
            [slug] =&gt; current state representation
            [parent] =&gt; 
            [name] =&gt; current state representation
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; In {{robotics}} a current state representation records only a single interpretaion or modle of the world that is incrementally updated with each new observation.  This requires less resources that a historical representation, which keeps track of the hstory of observationsn and actions, but isunable to re-assess past observations given new evidence.
        )

    [cyberattack] =&gt; Array
        (
            [slug] =&gt; cyberattack
            [parent] =&gt; 
            [name] =&gt; cyberattack
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; A cyberattack is when digital means are used to infiltrate or damage a compter system.  This may involve some form of computer virus, worm or other self-proposgating code.  It may also involve more human attacks, such as {{phishing}}.  Cyberattacks need not compromise a compter sysyem to casuse damage, in particular {{denial of service}} ({{DoS}}) atacks simply flood a machine with a vast  level of network traffic. the   The outcome of a cyber attack might be to bring donw the system, to extact information, or to leave an infiltaretd system avaiulable to itself perform attacks.  For example, {{IOT}} devices typically have poor security and may be targetted, not to daage them per se, but so that they can be part of a {{bot army}} for {{DoS}} attacks on the real targets.
        )

    [cyberattacks] =&gt; Array
        (
            [slug] =&gt; cyberattacks
            [parent] =&gt; 
            [name] =&gt; cyberattacks
            [canonical] =&gt; cyberattack
            [see] =&gt; 
            [description] =&gt; 
        )

    [cyberwarfare] =&gt; Array
        (
            [slug] =&gt; cyberwarfare
            [parent] =&gt; 
            [name] =&gt; cyberwarfare
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; Cyberwarfare is when {[cyber weapons}} or {{cyberattacks}} are used as part of allout war or so more limited conflicts.   More broadly AI and digital technology may be used as part of strategic lanning (such as Isreal's 'Gospel' system) or as part of coventional wepaons (such as GPS-guided smart bombs).   There is great concern about fully {{autonomous weapons}} that make their own decsions about what to attack (and who to kill) without a {{human in the loop}}.
        )

    [cyberweapons] =&gt; Array
        (
            [slug] =&gt; cyberweapons
            [parent] =&gt; 
            [name] =&gt; cyberweapons
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; Cyberweapons is the name of a computational algorithm or device  that can be used to inflict damage.  them ipact may be purely in the digital realm, for examplel, disrupting communications, but may also target physical infrastructure.  Probably the first example of the former was {{Stuxnet}}, which, in the late 2000s, targeted Iranian centrifuges used for nuclear processing.  
        )

    [cyc] =&gt; Array
        (
            [slug] =&gt; cyc
            [parent] =&gt; 
            [name] =&gt; CYC
            [canonical] =&gt; cyc project
            [see] =&gt; 
            [description] =&gt; 
        )

    [cyc project] =&gt; Array
        (
            [slug] =&gt; cyc project
            [parent] =&gt; 
            [name] =&gt; CYC Project
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; CYC is a long running project to capture and reason about {{general knowledge}}. The CYC project was founded in 1984 by {{Doug Lenat}}.  It uses {{symbolic AI}} explicitly representing knowledge and reasoning and it is the largest endeavour of its kind: OpenCYC 4.0, released in 2012, contains around a quarter of a million concepts and over two million facts about them ins a large {{semantic network}}.  Although it is a general knowledge base, it is divided into microtheories about different topic areas, this structure has enabled some of its successful applications in specific sub-areas.   In some ways {{large-language models}} capture elements of general knowledge due to the vast text sources they process, but lack the explicit {{semantics}} and {{explainability}} of CYC.  In his last paper before his death, Lenat and long-term colleague Gary Marcus, wrote about the ptential of {{LLM}}s to learn frm CYC, in a form of {{neurosymbolic}} AI.
        )

    [cyclotorsion] =&gt; Array
        (
            [slug] =&gt; cyclotorsion
            [parent] =&gt; 
            [name] =&gt; cyclotorsion
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; In a camera, cyclotorsion is the ability to rotate the camera and thus the horizon
        )

    [daleks] =&gt; Array
        (
            [slug] =&gt; daleks
            [parent] =&gt; 
            [name] =&gt; Daleks
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; Daleks are a {{robot}}-like enemy of {Dr Who}}.  They have mechanical arms and an eye stalk, but are actually not a robot as such, but a form of prosthesis or {{cyborg}} as there is a creature within.
        )

    [damasio] =&gt; Array
        (
            [slug] =&gt; damasio
            [parent] =&gt; 
            [name] =&gt; damasio
            [canonical] =&gt; damasio, antonio
            [see] =&gt; 
            [description] =&gt; 
        )

    [damasio, antonio] =&gt; Array
        (
            [slug] =&gt; damasio, antonio
            [parent] =&gt; 
            [name] =&gt; Damasio, Antonio
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; Antonio Damasio is a neuroscientist and philosopher of consciousness.  His books include &quot;Descartes Error&quot;&quot;, which argues, inter alia, that the source of consciousness is knolwedge of one's internal bodily state.  Crucially, he challenges Descrates' separation of body and mind and argues that {{emotion}} and {{rationality}} are intinately linked.  This is important for AI research as it suggests that not only that {{artificial consciousness}}, if possible, depends on {{artificial emotion}}, but actually any true AI requires this too.
        )

    [dartmouth conference] =&gt; Array
        (
            [slug] =&gt; dartmouth conference
            [parent] =&gt; 
            [name] =&gt; Dartmouth Conference
            [canonical] =&gt; 
            [see] =&gt; dartmouth workshop
            [description] =&gt; 
        )

    [dartmouth workshop] =&gt; Array
        (
            [slug] =&gt; dartmouth workshop
            [parent] =&gt; 
            [name] =&gt; Dartmouth Workshop
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; The Dartmouth Workshop was a meeting of about a dozen researchers in 1956 at Dartmouth College.  It was here that the term 'Artificial Intelligence' was coined, and hence is the official starting point for the discipline, although, the roots cna be traced before this, not least amongst the attendees.
        )

    [data-driven reasoning] =&gt; Array
        (
            [slug] =&gt; data-driven reasoning
            [parent] =&gt; 
            [name] =&gt; data-driven reasoning
            [canonical] =&gt; 
            [see] =&gt; forward reasoning
            [description] =&gt; 
        )

    [data cleaning] =&gt; Array
        (
            [slug] =&gt; data cleaning
            [parent] =&gt; 
            [name] =&gt; data cleaning
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; 
        )

    [data detectors] =&gt; Array
        (
            [slug] =&gt; data detectors
            [parent] =&gt; 
            [name] =&gt; data detectors
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; 
        )

    [data documentation] =&gt; Array
        (
            [slug] =&gt; data documentation
            [parent] =&gt; 
            [name] =&gt; data documentation
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; 
        )

    [data fusion] =&gt; Array
        (
            [slug] =&gt; data fusion
            [parent] =&gt; 
            [name] =&gt; data fusion
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; 
        )

    [data reduction] =&gt; Array
        (
            [slug] =&gt; data reduction
            [parent] =&gt; 
            [name] =&gt; data reduction
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; 
        )

    [data structure] =&gt; Array
        (
            [slug] =&gt; data structure
            [parent] =&gt; 
            [name] =&gt; data structure
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; 
        )

    [data wrangling] =&gt; Array
        (
            [slug] =&gt; data wrangling
            [parent] =&gt; 
            [name] =&gt; data wrangling
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; 
        )

    [database] =&gt; Array
        (
            [slug] =&gt; database
            [parent] =&gt; 
            [name] =&gt; database
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; 
        )

    [de-bias] =&gt; Array
        (
            [slug] =&gt; de-bias
            [parent] =&gt; 
            [name] =&gt; de-bias
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; 
        )

    [de-trending] =&gt; Array
        (
            [slug] =&gt; de-trending
            [parent] =&gt; 
            [name] =&gt; de-trending
            [canonical] =&gt; 
            [see] =&gt; trend removal
            [description] =&gt; 
        )

    [decibel] =&gt; Array
        (
            [slug] =&gt; decibel
            [parent] =&gt; 
            [name] =&gt; decibel
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; The decibel scale is used to measure sound levels.   It is a {{logarithmic scale}} where an increase in 10 decibels means the sound is 10 times louder (as measured by sound energy).  This means that a 20 decibel difference represents a 100 times louder, etc.  The {{logarithmc scale}} is used becasue sounds differ greatly n lodness. For example, a whisper is about 30db, conversation about 60db (one thousand times louder) and nearby car engine 90db (one million times loder than the whisper!).
        )

    [decision support system] =&gt; Array
        (
            [slug] =&gt; decision support system
            [parent] =&gt; 
            [name] =&gt; decision support system
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; 
        )

    [decision support systems] =&gt; Array
        (
            [slug] =&gt; decision support systems
            [parent] =&gt; 
            [name] =&gt; decision support systems
            [canonical] =&gt; decision support system
            [see] =&gt; 
            [description] =&gt; 
        )

    [decision tree] =&gt; Array
        (
            [slug] =&gt; decision tree
            [parent] =&gt; 
            [name] =&gt; decision tree
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; 
        )

    [decision trees] =&gt; Array
        (
            [slug] =&gt; decision trees
            [parent] =&gt; 
            [name] =&gt; decision trees
            [canonical] =&gt; 
            [see] =&gt; decision tree
            [description] =&gt; 
        )

    [deduction] =&gt; Array
        (
            [slug] =&gt; deduction
            [parent] =&gt; 
            [name] =&gt; deduction
            [canonical] =&gt; 
            [see] =&gt; deductive reasoning
            [description] =&gt; 
        )

    [declarative knowledge] =&gt; Array
        (
            [slug] =&gt; declarative knowledge
            [parent] =&gt; 
            [name] =&gt; declarative knowledge
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; Declarative knowledge is about rememebered facts about the world, in contrast to {{procedural knowledge}}, which tries tom work out things on the fly.  This distinction applies to {{human reasoning}}, for example, one perosn might recall their seven times table (decarative knowledge), wheras another might recall one or two key multiples and then rapdily add or subtract sevens ({{procedural knowledge}}).  It also applies to {{automated reasoning}} .  In particular {{machine learning}} and {{statistical algorithms}} using the web and other {{big data}} techniques allow declarative reasoning  that may appear intelligent without any real understanding.
        )

    [deductive learning] =&gt; Array
        (
            [slug] =&gt; deductive learning
            [parent] =&gt; 
            [name] =&gt; deductive learning
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; 
        )

    [deductive reasoning] =&gt; Array
        (
            [slug] =&gt; deductive reasoning
            [parent] =&gt; 
            [name] =&gt; deductive reasoning
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; Deductive reasoning makes set of rules and known facts to deduce further knowledge.  For example, if you know that grass is green and a field is full of grass, you may deduce that a field will be green.  Deductive reasoning is usually limited to what is definitely known, as opposed to other forms of reasoning, such as {{abduction}}, which may reason from {{incomplete knowledege}}. 
        )

    [deep blue] =&gt; Array
        (
            [slug] =&gt; deep blue
            [parent] =&gt; 
            [name] =&gt; Deep Blue
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; Deep Blue is the {{IBM}}  {{chess computer}} that defeated the chess grandmaster Garry Kasparov in {{1996}}.
        )

    [deep fakes] =&gt; Array
        (
            [slug] =&gt; deep fakes
            [parent] =&gt; 
            [name] =&gt; deep fakes
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; 
        )

    [deep learning] =&gt; Array
        (
            [slug] =&gt; deep learning
            [parent] =&gt; 
            [name] =&gt; deep learning
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; 
        )

    [deep neural network] =&gt; Array
        (
            [slug] =&gt; deep neural network
            [parent] =&gt; 
            [name] =&gt; deep neural network
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; A deep neural network is a {{neural network}} with lots of layers.. Tyoically the layers are also large, in the sense of lots of nodes.  Each layer can also have a differnt size and use different types of learning.  For example, it is common for the first layer to be a {{Restricted Boltzmann Machine}} in order to perofrma {{dimensional reduction}}.  TYpically the inner laters are {{underdetermined}} (many equally good arrangements of weightings); this and theor disrance from the output layer means that {{backpropogation}} or similar learning rules need to have very slow learning rates to avoid instabilities.  In addition, more layers and laregr layers usually req=uires more tarng data.  Together these mean that {{deep learning}} (training of deep neural networks) requires. a lot of training data.  This combination of computational cost and data volume is one of the main reasons that the use deep neural networks were not widely adopted for many years.
        )

    [deep neural networks] =&gt; Array
        (
            [slug] =&gt; deep neural networks
            [parent] =&gt; 
            [name] =&gt; deep neural networks
            [canonical] =&gt; deep neural network
            [see] =&gt; 
            [description] =&gt; 
        )

    [definite clause grammar] =&gt; Array
        (
            [slug] =&gt; definite clause grammar
            [parent] =&gt; 
            [name] =&gt; definite clause grammar
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; A definite clause grammar is a way of representing grammars as {{definite clauss}} in {{Prolog}} and related languages.  This allows the grammar to express a level of semantics in the form of 'features' and conditions on these, thus adding a level of {[semantcis}} to the grammar.
        )

    [degrees of freedom (data)] =&gt; Array
        (
            [slug] =&gt; degrees of freedom (data)
            [parent] =&gt; 
            [name] =&gt; degrees of freedom (data)
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; 
        )

    [degrees of freedom (robotics)] =&gt; Array
        (
            [slug] =&gt; degrees of freedom (robotics)
            [parent] =&gt; 
            [name] =&gt; degrees of freedom (robotics)
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; 
        )

    [delta] =&gt; Array
        (
            [slug] =&gt; delta
            [parent] =&gt; 
            [name] =&gt; delta
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; A delta is a small change.  For files it is a description of the differences between the files and is often used in {{version control}} systenms.  In {{incremental algorithms}} it is the small change in parameter values at each step, for example the updates to weights in {{backpropagation}}.
        )

    [deltas] =&gt; Array
        (
            [slug] =&gt; deltas
            [parent] =&gt; 
            [name] =&gt; deltas
            [canonical] =&gt; delta
            [see] =&gt; 
            [description] =&gt; 
        )

    [dendral] =&gt; Array
        (
            [slug] =&gt; dendral
            [parent] =&gt; 
            [name] =&gt; DENDRAL
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; DENDRAL was an early {{expert system}}, developed initially in the 1960s, that used {{domain specific knowledge}} about chemical structure in order to work alongside chemists to create new scientifc results.  In some ways this can be seen as a precursor to current scientific discovery systems such as {{AlphaFold}}.
        )

    [denial of service (dos)] =&gt; Array
        (
            [slug] =&gt; denial of service (dos)
            [parent] =&gt; 
            [name] =&gt; denial of service (DoS)
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; A denial of service (DoS) s a form of {{cyberattack}} where large numbers of computers all tryn to access the same computer, often a {{web server}}, menaing that eiether it crashes, or is so slow to respond that legitimate traffic cannot access it.  Typically, the computers performing the atack are themselves machies that have been compromised through a computer virus.
        )

    [dependent feature] =&gt; Array
        (
            [slug] =&gt; dependent feature
            [parent] =&gt; 
            [name] =&gt; dependent feature
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; A dependent feature is a feature that only makes sense when some other feature has a specific value.  For example, in a {{database}} of motorvehicles 'number fo doors' is not meaningful if the type of vehicle is 'motorcycle'.  Many {macine learning}} algorithms will perform more efficiently if they take into account this form of dependency, for exmaple, in a {{genetic algorithm}} it is not helpful to {{mutate}} features that cannot have any effect.
        )

    [dependent variable] =&gt; Array
        (
            [slug] =&gt; dependent variable
            [parent] =&gt; 
            [name] =&gt; dependent variable
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; In {{linear regression}} and other foms of modelling a {{dependent variable}} is a value that you cannot control or you wish to have as the output of the model.  In contrast to an {{independent variable}}, which is something you can modify or one of the model's inputs.  That is the model is effectively of the form: &lt;br&gt; &amp;nbsp; &amp;nbsp; &amp;nbsp; dependent variables &amp;nbsp; = &amp;nbsp;  model( parameters, independent variables ) &lt;br&gt; where one then tries to fit the parameters.
        )

    [dependent variables] =&gt; Array
        (
            [slug] =&gt; dependent variables
            [parent] =&gt; 
            [name] =&gt; dependent variables
            [canonical] =&gt; dependent variable
            [see] =&gt; 
            [description] =&gt; 
        )

    [depth first] =&gt; Array
        (
            [slug] =&gt; depth first
            [parent] =&gt; 
            [name] =&gt; depth first
            [canonical] =&gt; depth first search
            [see] =&gt; 
            [description] =&gt; 
        )

    [depth first search] =&gt; Array
        (
            [slug] =&gt; depth first search
            [parent] =&gt; 
            [name] =&gt; depth first search
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; 
        )

    [depth of search tree] =&gt; Array
        (
            [slug] =&gt; depth of search tree
            [parent] =&gt; 
            [name] =&gt; depth of search tree
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; The depth of a {{search tree}} is the number of links from the root to the deepest leaf node.
        )

    [derivational analogy] =&gt; Array
        (
            [slug] =&gt; derivational analogy
            [parent] =&gt; 
            [name] =&gt; derivational analogy
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; A derivational analogy is a form of {{analogy}} where the process by which an analogous solution was found is used as well as the solution itself.   This allows new solutions that exploit the way the previous analogous solutuon was found, but may vary the steps.  The new solution iself may look nothing like the old one (taking the {{analogy}} into account), but the solution path does.  For example, soemone might reason &quot;getting out of an escape room is a bit like a maths exam question&quot;.  To solve the maths question they would first of all list all the information given, so in the escape room, they list all the materials that are available.  The solutons for the maths question and the escape room are nothing like each other, but the processes are similar.

        )

    [descartes] =&gt; Array
        (
            [slug] =&gt; descartes
            [parent] =&gt; 
            [name] =&gt; Descartes
            [canonical] =&gt; descartes, rene
            [see] =&gt; 
            [description] =&gt; 
        )

    [descartes, rene] =&gt; Array
        (
            [slug] =&gt; descartes, rene
            [parent] =&gt; 
            [name] =&gt; Descartes, R&amp;eacute;ne
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; R&amp;eacute;ne Descartes (1596&ndash;1650) was a matehmatician and philosopher famous, amngst other things, for his statement &quot;I think therefore I am&quot;.  His work on geometry led to what we now call {{Cartesian coordinates}} which are used universally in {{graphics}}, {{computer vision}} and {{GIS}}.  He believed that there was a sharp divide between {res cogitans}}, the material that forms mind, and {{res extensa}}, the physical body -- a form of {{dualism}}.  This view is near universally rejected by modern philosophers of {{intelligence}} and {{conscousness}}, who instead espouse {{physicalism}} the belief that all thought and human subjectuve experience cam be explained through physical phenomena.
        )

    [deterministic] =&gt; Array
        (
            [slug] =&gt; deterministic
            [parent] =&gt; 
            [name] =&gt; deterministic
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; Deterministic phenomena are ones which follow a fixed set of physical or logical rules.  The same initial state will always lead to the same outcomes. In contrast, {{stochastic}} phenomen have some degree of {{random}}, {{probabalistic}} or {{nondeterminsitic}} behaviour. 
        )

    [deterministic ground] =&gt; Array
        (
            [slug] =&gt; deterministic ground
            [parent] =&gt; 
            [name] =&gt; deterministic ground
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; In an {{adaptive user interface}}, determinsitic ground refers to the thinsg tht are not adapted by the system, and therefore can be relied upon to have consistent behaviour.  For example, an adaptive website might have a recommended path suggesting the next page to look at (adaptive), but also have a fixed navigation menu (deterministic ground).
        )

    [deterministic search] =&gt; Array
        (
            [slug] =&gt; deterministic search
            [parent] =&gt; 
            [name] =&gt; deterministic search
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; A determinsitic search uses a fixed rule to decide the order in which to visit states in a {{search tree} or {{search graph}}, and thus will always search in the same order given the same data.  In contrast a {{stochastic search}} may take different choices based on {{random}} chance.
        )

    [dialogue component] =&gt; Array
        (
            [slug] =&gt; dialogue component
            [parent] =&gt; 
            [name] =&gt; dialogue component
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; The dialogue component of an {{expert system}} is the part that interacts with the user.  This may be in the form of a question and answer dialogue, or may be more tabular, form-based or graphical.
        )

    [dialogue] =&gt; Array
        (
            [slug] =&gt; dialogue
            [parent] =&gt; 
            [name] =&gt; dialogue
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; The term 'dialogue' in human&ndash;computer interaction refers to the style and ordering of human and comoyter imouts and outputs.  This may be of a linguistic form, such as in a {{chatbot}}, but the term also includes menu-based, pointing, and graohical interactions.
            [subentries] =&gt; Array
                (
                    [dialogue!mixed control] =&gt; Array
                        (
                            [slug] =&gt; dialogue!mixed control
                            [parent] =&gt; dialogue
                            [name] =&gt; mixed control
                            [canonical] =&gt; 
                            [see] =&gt; 
                            [description] =&gt; In an {{expert system}}, mixed control is when both user and computer system can initiate interactions.  For example, the user might ask questions of the {{expert system}}, but the system might then ask for information it needs or proactively {{notify}} the user when it notices and important change.
                        )

                    [dialogue!system control] =&gt; Array
                        (
                            [slug] =&gt; dialogue!system control
                            [parent] =&gt; dialogue
                            [name] =&gt; system control
                            [canonical] =&gt; 
                            [see] =&gt; 
                            [description] =&gt; In an {{expert system}}, system control is when the computer system drives the interaction.  The user may supply input, but only when asked.
                        )

                    [dialogue!user control] =&gt; Array
                        (
                            [slug] =&gt; dialogue!user control
                            [parent] =&gt; dialogue
                            [name] =&gt; user control
                            [canonical] =&gt; 
                            [see] =&gt; 
                            [description] =&gt; In an {{expert system}}, user control is when the user drives interactions.  The system can only respond to questions or actions initiated by the {{user}}.
                        )

                )

        )

    [dialogue!mixed control] =&gt; Array
        (
            [slug] =&gt; dialogue!mixed control
            [parent] =&gt; dialogue
            [name] =&gt; mixed control
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; In an {{expert system}}, mixed control is when both user and computer system can initiate interactions.  For example, the user might ask questions of the {{expert system}}, but the system might then ask for information it needs or proactively {{notify}} the user when it notices and important change.
        )

    [dialogue!system control] =&gt; Array
        (
            [slug] =&gt; dialogue!system control
            [parent] =&gt; dialogue
            [name] =&gt; system control
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; In an {{expert system}}, system control is when the computer system drives the interaction.  The user may supply input, but only when asked.
        )

    [dialogue!user control] =&gt; Array
        (
            [slug] =&gt; dialogue!user control
            [parent] =&gt; dialogue
            [name] =&gt; user control
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; In an {{expert system}}, user control is when the user drives interactions.  The system can only respond to questions or actions initiated by the {{user}}.
        )

    [difference engine] =&gt; Array
        (
            [slug] =&gt; difference engine
            [parent] =&gt; 
            [name] =&gt; Difference Engine
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; The Difference Engine was an early computer designed by {{Charles Babbage}} in the first half of the 19th Century.  It uses the method of 'finite differences' to calcualte polynomials, and is thus a fixed purpose calculator unlike Babbage's later designs for the {{Analytic Engine}}, the worlds first programmable computer, alas never constructed except in steam-punk fiction.
        )

    [differential] =&gt; Array
        (
            [slug] =&gt; differential
            [parent] =&gt; 
            [name] =&gt; differential
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; 
        )

    [diffusion models] =&gt; Array
        (
            [slug] =&gt; diffusion models
            [parent] =&gt; 
            [name] =&gt; diffusion models
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; 
        )

    [digital filtering] =&gt; Array
        (
            [slug] =&gt; digital filtering
            [parent] =&gt; 
            [name] =&gt; digital filtering
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; 
        )

    [digital signal processing] =&gt; Array
        (
            [slug] =&gt; digital signal processing
            [parent] =&gt; 
            [name] =&gt; digital signal processing
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; Digital signal processing (DSP) refers to the computer processing of time-series data such as radio or sound.  It is in contrast to various forms of analogue signal processing, involving circuits with fixed components such as resistors and transistors.  Some forms of DSP effectively emulate the largely linear processing of theor analogue predecessors, band-pass filters that tune a singal to a particular frequency.  However, it is also possible to produce more complex filters, for example those that speed up speech without increasing the pitch.
        )

    [digitization] =&gt; Array
        (
            [slug] =&gt; digitization
            [parent] =&gt; 
            [name] =&gt; digitization
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; 
        )

    [dimension reduction] =&gt; Array
        (
            [slug] =&gt; dimension reduction
            [parent] =&gt; 
            [name] =&gt; dimension reduction
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; 
        )

    [dimensional reduction] =&gt; Array
        (
            [slug] =&gt; dimensional reduction
            [parent] =&gt; 
            [name] =&gt; dimensional reduction
            [canonical] =&gt; dimension reduction
            [see] =&gt; 
            [description] =&gt; 
        )

    [dimensionality reduction] =&gt; Array
        (
            [slug] =&gt; dimensionality reduction
            [parent] =&gt; 
            [name] =&gt; dimensionality reduction
            [canonical] =&gt; dimension reduction
            [see] =&gt; 
            [description] =&gt; 
        )

    [directed graph] =&gt; Array
        (
            [slug] =&gt; directed graph
            [parent] =&gt; 
            [name] =&gt; directed graph
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; A directed graph is a network of nodes and connections (often called edges), where each edge has a direction.  This is rather like a city with a one-way system.  For example, a graph with three nodes A, B and C, might have the following dirceted edges: A&amp;rarr;B, B&amp;rarr;A and B&amp;rarr;C.  Note how there are links between A and B in both directions; there is a link going from B to C, but not the other way round; and there is no link at all between A and C.  Directed graphs often arise in {{games}}  where moves take one from one gae state to anther, but often are not symetric; so that you can't always reverse your last move.&lt;br&gt;&lt;br&gt; A directed graph is in contrast to an {{undirected graph}} where edges have no direction and effectively link both ways.
        )

    [disambiguation] =&gt; Array
        (
            [slug] =&gt; disambiguation
            [parent] =&gt; 
            [name] =&gt; disambiguation
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; 
        )

    [discrete distributions] =&gt; Array
        (
            [slug] =&gt; discrete distributions
            [parent] =&gt; 
            [name] =&gt; discrete distributions
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; A discrete distrbution is a probability distribution where the underlying data are {{discrete}} such as integers or a finite set of values.  Examples include the probabilities of different eye colours in a population ({{categorical}} underlying data), or the score after trou=wing a pair of dice ({{numerical}} underlying data).
        )

    [distributed ai] =&gt; Array
        (
            [slug] =&gt; distributed ai
            [parent] =&gt; 
            [name] =&gt; distributed AI
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; Distributed AI is where several agents (robots or software), each with AI capability interact with one another communicating and/or acting in a shared environment.  This may involve collaboratively solving a problem such as {{factory robots}} reconfiguring to deal with changing orders or faults.  Alternatively may be a looser form of interaction such as multiple {{autonomous vehicles}} in a busy urban environment.  Note the term is not usually applied when a single AI algorithm is employs {{parallel processing}}, not for {{ensemble methods}} or {{swarm computing}} employing vast numbers of simple agents.
        )

    [distributed cognition] =&gt; Array
        (
            [slug] =&gt; distributed cognition
            [parent] =&gt; 
            [name] =&gt; distributed cognition
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; Distributed cognition refers to the way that humans (and other animals) effectibely think using aspects of their enviornment inclduing interactions with other people.  Ths may be explciit, such as using a calcuator to help arithmetic.  It may also be implicit such as knowing were one is in the process of assembling a piece of flat-pack furnture becasue you can see the part-assembled piece in front of you.  Aspects of distributed cognition can form the inspiration for {{distributed AI}} and also for understanding the behaviour of {{hybrid human-AI}} systems.
        )

    [divide and conquer] =&gt; Array
        (
            [slug] =&gt; divide and conquer
            [parent] =&gt; 
            [name] =&gt; divide and conquer
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; Divide and conquer is a general {{problem solving}} technique to take a large prolem and braek it into smaller pieces, ecah of which is then easier to solve.  In computer algorithms, the same divide and conquer approach is usually performed {{recursively}} on the parts.  In a {{distributed computing}} setting, such as {{MapReduce}}, different processors may attack different sub-problems.  In an {{agent}}-base dplatform, different agents may divide up a problem beased on their expertise.
        )

    [document retrieval] =&gt; Array
        (
            [slug] =&gt; document retrieval
            [parent] =&gt; 
            [name] =&gt; document retrieval
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; Document retrieval (or {{text retreival}} refers to the algorithms and methods used ot query and loacte doucments in large information systems.  In a structured database queries often use indices for specific keys, but within a document retreival system a {{free text index}} is usually required.  More complex {{text processing}} may be used to computer similarities between documents or looking at human usage patterns for {{recommender systems}} to suggest potential documents.
        )

    [domain-independent knowledge] =&gt; Array
        (
            [slug] =&gt; domain-independent knowledge
            [parent] =&gt; 
            [name] =&gt; domain-independent knowledge
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; Domain-independent knowledge is the kind of {{common sense}} or {{general knowledge}} that one has without studying a specific topic or area of expertise.  Simple domain-independent knowledge, such as the idea of generalising firm past examples, may be implicitly built inot the way algoritms work.  More extensive domain-independent knowledge, such as the fact that cats chase birds, is far harder for machines, yet simple for humans.  The long standing {{CYC}} project is one attempt to deal with ths and one can arge the {{large language models}} do capture some level of {{general knowledge}} albiet implicitly in the {{deep neural network}} weights.  Domain-independent knowledge is used in contrast with {{domain-dependent knowledge}}, the specialised knowkedge of a specific area. 
        )

    [domain-specific knowledge] =&gt; Array
        (
            [slug] =&gt; domain-specific knowledge
            [parent] =&gt; 
            [name] =&gt; domain-specific knowledge
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; Domain-dependent knowledge is the specialised knowkedge of a specific topic or area of expertise.     Domain-dependent knowledge is used in contrast with {{domain-independent knowledge}}, which is {{common sense}} or {{general knowledge}}.  For example, a medical {{expert system}} may use a lareg mdeical {{ontology}} or use data on the prevalance of diseases to perform {{Bayesian reasonong}}.  While the rules of reasoning are a simple form of {{domain-independent knowledge}}, the {{ontology}} and detailed daat are domain dependent.
        )

    [dominoes] =&gt; Array
        (
            [slug] =&gt; dominoes
            [parent] =&gt; 
            [name] =&gt; dominoes
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; Dominoes is the name of the small tiles used in a table top game and of the game itself.  Each domino usually has between zero and six dots at either end, but there are sets with more.  In the AI Book they are used in a simple board game {{placing dominoes}}, which ignores the numbers on the tiles, to illustrate {{game trees}}
        )

    [dot-com period] =&gt; Array
        (
            [slug] =&gt; dot-com period
            [parent] =&gt; 
            [name] =&gt; dot-com period
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; The dot-com years refers to the period from approximately late 1998 to early 2000  when there was an explosion of web start-ups and venture funding for commercial hi-tech activity.  It was fuelled by the increasing web user base, but for a period hype overtook the human and technical foundations leading, perhaps inevitably, to a major crash, 'the dot-com bubble'.  The latter was triggerd by the flotation of lastminute.com -- which paradoxically is still a major interbet travel company.  The share price of lastminute rose sharply after listing, but this was soon followed by an even sharper drop.  Confidence in hi-tech companies was hit leading to a near shotdown of funding for hi-tech investments that lasted for several years. 
        )

    [down-sample] =&gt; Array
        (
            [slug] =&gt; down-sample
            [parent] =&gt; 
            [name] =&gt; down-sample
            [canonical] =&gt; down-sampling
            [see] =&gt; 
            [description] =&gt; 
        )

    [down-sampling] =&gt; Array
        (
            [slug] =&gt; down-sampling
            [parent] =&gt; 
            [name] =&gt; down-sampling
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; Down sampling is a term from {{signal processing}} and {{time-series analysis}}.  It refers to reducing the sampling points, often by simply picking every n-th sample.  For example, given the 20 second sampled data: &lt;br&gt; &amp;nbsp; &amp;nbsp; &amp;nbsp; [1.35, 1.37, 1.38, 1.45, 1.42, 1.48, 1.52, 1.51, 1.56], &lt;br&gt;we might take every 3rd item to give one minute samples [1.35, 1.45, 1.52].  More sophisticated down sampling might do some form of {{moving average}} or other {{smoothing}} process before the selection, for example, taking one munute running averages: [1.36, 1.45, 1.53].  &lt;br&gt;&lt;br&gt;Downsamopling is particularly useful as a {{data reduction}} step for slowly changing data.  In a {{window-based algorithm}} it effectively increases how far back the algorithm can look for a given window size.
        )

    [dreyfus, h] =&gt; Array
        (
            [slug] =&gt; dreyfus, h
            [parent] =&gt; 
            [name] =&gt; Dreyfus, H
            [canonical] =&gt; dreyfus, hubert
            [see] =&gt; 
            [description] =&gt; 
        )

    [dreyfus, hubert] =&gt; Array
        (
            [slug] =&gt; dreyfus, hubert
            [parent] =&gt; 
            [name] =&gt; Dreyfus, Hubert
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; Hubert Dreyfus was a philosopher who was part of the early discussions on the potential for true {{artificial itelligence}}.  He effectvely argued against {{strong AI}} 
 and the {{physical symbol system hypothesis}} suggesting instead that the highly contextual and embodied natiue of intelligence meant that there was a sharp divide between biological human intelligence and computation.
        )

    [drunkards walk] =&gt; Array
        (
            [slug] =&gt; drunkards walk
            [parent] =&gt; 
            [name] =&gt; drunkards walk
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; A drunkards walk refers to a {{stochasric process}} where each step is random, and not based on past history at all.  You can create a simple one-dimensional drunkard's walk by tossing a coin then moving forward a step for each head and back a step for each tail.
        )

    [dsp] =&gt; Array
        (
            [slug] =&gt; dsp
            [parent] =&gt; 
            [name] =&gt; DSP 
            [canonical] =&gt; 
            [see] =&gt; digital signal processing
            [description] =&gt; 
        )

    [dualism] =&gt; Array
        (
            [slug] =&gt; dualism
            [parent] =&gt; 
            [name] =&gt; dualism
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; 
        )

    [dynamic function allocation] =&gt; Array
        (
            [slug] =&gt; dynamic function allocation
            [parent] =&gt; 
            [name] =&gt; dynamic function allocation
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; 
        )

    [eager] =&gt; Array
        (
            [slug] =&gt; eager
            [parent] =&gt; 
            [name] =&gt; EAGER
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; Eager was an early {{intelligent agent}} that used {{programming-by-demonstration}} to automate user 
        )

    [ebl] =&gt; Array
        (
            [slug] =&gt; ebl
            [parent] =&gt; 
            [name] =&gt; EBL
            [canonical] =&gt; 
            [see] =&gt; explanation-based learning
            [description] =&gt; 
        )

    [ecg] =&gt; Array
        (
            [slug] =&gt; ecg
            [parent] =&gt; 
            [name] =&gt; ECG
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; An ECG (electo-cardiogram) is a device to measure electrical activity in the heart.  The output is the heart beat trace that you will have seen in medical dramas.   Analysing an ECG is challenging for time-series methods as it consist of bursts of actuvity (the heart beat) where the shape of each burst and the spacing between them are each important.  Furthernore the detailed shape of each burst and the length also vary.  Traditional {{time-series analysis}} methods prefer data that is more uniform in structure (even if stochastic), as do some {{neural metods}}.  Some form of multi-level analysis is typically required either explicitly by applying different methods to separate the trace into components and then analyse each, or implicitly, for example in the layers of a {{deep neural network}}.      &lt;br&gt; &lt;a href=&quot;https://commons.wikimedia.org/wiki/File:12-lead_ECG.jpg&quot; target=&quot;_blank&quot;&gt;&lt;img align=&quot;center&quot; src=&quot;https://en.wikipedia.org/wiki/Electrocardiography#/media/File:12-lead_ECG.jpg&quot;&gt;&lt;/a&gt;
        )

    [edge detection] =&gt; Array
        (
            [slug] =&gt; edge detection
            [parent] =&gt; 
            [name] =&gt; edge detection
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; 
        )

    [edge following] =&gt; Array
        (
            [slug] =&gt; edge following
            [parent] =&gt; 
            [name] =&gt; edge following
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; 
        )

    [egocentrism] =&gt; Array
        (
            [slug] =&gt; egocentrism
            [parent] =&gt; 
            [name] =&gt; egocentrism
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; This refers to the way that young children, and also adults, see thew world from their own perspective and may find it hard even to realise that the world can be seen diffeerntly.  An extreme example of this is the child's game of peek-a-boo, where a young child believes themselves to not be see-able if they can't see.   {{Piaget}} measured this effect by creating a landscape and asking chidlren what could be seen from alocation; young children say that something can be seen if they can see it, whereas older chidlren are able to realise that one side of a hill is not visible from the other.  Although adults are unlikely the think like this at the level of locations and seeing, we still may find it hard to understand that people think differently to us or have fundamentally different viewpoints or values.
        )

    [eigenvalues] =&gt; Array
        (
            [slug] =&gt; eigenvalues
            [parent] =&gt; 
            [name] =&gt; eigenvalues
            [canonical] =&gt; 
            [see] =&gt; eigenvector
            [description] =&gt; 
        )

    [eigenvector] =&gt; Array
        (
            [slug] =&gt; eigenvector
            [parent] =&gt; 
            [name] =&gt; eigenvector
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; An eigenvector of.a {{square matrix}} is a {{vector}} which when multiplied by the matrix ends up as a multiple of itself.  In otherwords, the eigenvector is unchanged except for scale.  Mathentaically, if M is the {{matrix}} and v an eignevector,  Mv = &amp;lambda;v.   The scale factor &amp;lambda; is known as the {{eigenvalue}} of the eigenvector.  Eignevectors have many uses, both directly represented within algorithsm, or as a way to epxlian other processes.  As an an example of the former, if a {{cross-correlation matrix}} is calclated for a dataset, and the eigenvectors with large {{eigenvalues}} computed, these correspond to the directions in which the data has greatest variation, a process called {{principle components analysis}}.  As an example of the altter, Google PageRank agorithm chaxcses link spreading a measure of importance form page to page, however this effectively computes the {{principal eigenvector}} (the one with largest {[eigenvalue}}) of the {{web graph}} represented as a {{connection matrix}}.
        )

    [eigenvectors] =&gt; Array
        (
            [slug] =&gt; eigenvectors
            [parent] =&gt; 
            [name] =&gt; eigenvectors
            [canonical] =&gt; eigenvector
            [see] =&gt; 
            [description] =&gt; 
        )

    [eight queens problem] =&gt; Array
        (
            [slug] =&gt; eight queens problem
            [parent] =&gt; 
            [name] =&gt; eight queens problem
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; The eight queens problem tries to place eight queens on a standard 8x8 chess board so that no two queens are on the same row, column or diagonal.,
        )

    [eliza] =&gt; Array
        (
            [slug] =&gt; eliza
            [parent] =&gt; 
            [name] =&gt; ELIZA
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; ELIZA was an early {{chatbot}} created by{{ Joseph Weizenbaum}}.  It used simple {{template matching}} together with some fixed  prompts which it applied in a semi-random fashion.  The most well known variatnt was based on Rogerian psychology, where the human therapist is trained to relfect back the patient's statements.  For example, if ELIZA detected key terms such as 'mother' in the user's input, it might say &quot;tell me more about your mother&quot;.
        )

    [email filtering agents] =&gt; Array
        (
            [slug] =&gt; email filtering agents
            [parent] =&gt; 
            [name] =&gt; email filtering agents
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; 
        )

    [embodied] =&gt; Array
        (
            [slug] =&gt; embodied
            [parent] =&gt; 
            [name] =&gt; embodied
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; 
        )

    [emergent behaviour] =&gt; Array
        (
            [slug] =&gt; emergent behaviour
            [parent] =&gt; 
            [name] =&gt; emergent behaviour
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; 
        )

    [emotion] =&gt; Array
        (
            [slug] =&gt; emotion
            [parent] =&gt; 
            [name] =&gt; emotion
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; In day-to-day language, emotion has a broad menaing encomasing all of the ways we assess and feel about the world and peope and thinsg in it:  love, fear, joy, surprise.   In {{psychology}} emotion refers to physiological effects such heart rate, whereas {{feeling}} refers to the interpretation of this in higher-order ways such as 'fear' or 'excitement'.  In addition, {{mood}} refers to the longer term orientation, for exampel, if you are feeling particularly down, or particularly positive a about things.  Crucially in many situations the physiological effect (emotion) preceeds the interpretation ({{feeling}}), and the latter deoends on your {{mood}}.  For example,  if someone burst a balloon your heart will instantly race, but whether this leads to a sense of fun, fear or anger depends on your {{mood} and the situation.   In day-to-day speech both emotion and feeling are often called 'emotion'.  The distinction is important, however, when creating {{intelligent user interfaces}} that seek to interpret or predict human emotional responses.  It can also be useful when considering {{artificial emotion}}, with different ways to emulate the physiological side and the interpretation.
        )

    [empathy] =&gt; Array
        (
            [slug] =&gt; empathy
            [parent] =&gt; 
            [name] =&gt; empathy
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; 
        )

    [energy landscape] =&gt; Array
        (
            [slug] =&gt; energy landscape
            [parent] =&gt; 
            [name] =&gt; energy landscape
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; 
        )

    [enigma machine] =&gt; Array
        (
            [slug] =&gt; enigma machine
            [parent] =&gt; 
            [name] =&gt; Enigma machine
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; The Enigma machine was an encryption device used by the German armed forces during the Second World War.  {{Alan Turing}} and others at Bletchley Park developed the {{Bombe}} a device to help decrypt Enigma machine messages.  This was one of the inspirations for his later work studying computation.
        )

    [ensemble algorithms] =&gt; Array
        (
            [slug] =&gt; ensemble algorithms
            [parent] =&gt; 
            [name] =&gt; ensemble algorithms
            [canonical] =&gt; ensemble methods
            [see] =&gt; 
            [description] =&gt; 
        )

    [ensemble methods] =&gt; Array
        (
            [slug] =&gt; ensemble methods
            [parent] =&gt; 
            [name] =&gt; ensemble methods
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; Ensemble methods use large numbers of simpler AI units to solve complex problems.  A common example is the {{random forest}} which combines the results of large numbers of {{decsion trees}}, but other kinds of algorithms can be used for the simpler units.  The combination method may be a simple vote, or may involve more complex weighting of the results of individual units.
            [subentries] =&gt; Array
                (
                    [ensemble methods!heterogeneous] =&gt; Array
                        (
                            [slug] =&gt; ensemble methods!heterogeneous
                            [parent] =&gt; ensemble methods
                            [name] =&gt; heterogeneous
                            [canonical] =&gt; 
                            [see] =&gt; 
                            [description] =&gt; 
                        )

                    [ensemble methods!homogeneous] =&gt; Array
                        (
                            [slug] =&gt; ensemble methods!homogeneous
                            [parent] =&gt; ensemble methods
                            [name] =&gt; homogeneous
                            [canonical] =&gt; 
                            [see] =&gt; 
                            [description] =&gt; 
                        )

                )

        )

    [ensemble methods!heterogeneous] =&gt; Array
        (
            [slug] =&gt; ensemble methods!heterogeneous
            [parent] =&gt; ensemble methods
            [name] =&gt; heterogeneous
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; 
        )

    [ensemble methods!homogeneous] =&gt; Array
        (
            [slug] =&gt; ensemble methods!homogeneous
            [parent] =&gt; ensemble methods
            [name] =&gt; homogeneous
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; 
        )

    [entity recognition] =&gt; Array
        (
            [slug] =&gt; entity recognition
            [parent] =&gt; 
            [name] =&gt; entity recognition
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; Entity recognition refers to the process of finding words, phrases or non-text patterns that represent a thing such as person, number or date.  It typically precedes other forms of {{document processingng}} or {{Intelligent user interfaces}}.  For example, {{data-detectors}} scan emails, web pages or other documents to highlight entities such as {{dates}} and offer shorcuts to things that you can do with them.    Some datatypes such as numbers and date will use some form of {{pattern matching}} such as {[regular expressions}}, others, such as placenames, may involve searching for words or n-grams in public or private databases, a process known as  {{named entity recognition}}.  Often simpler entities, may be used as part of higher-level entity detection; for example an email containing date, location and time could be recognised as refering to an event and thus offer the user the opportunity to add it to their calendar.
        )

    [entropy] =&gt; Array
        (
            [slug] =&gt; entropy
            [parent] =&gt; 
            [name] =&gt; entropy
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; Entropy is a mathematical measure of the {{information}} in a system.  For a single item of data with a number of possible values, each with {{probability}} p&lt;sub&gt;i&lt;/sub&gt;, the entrop is defined as follows:  &lt;br&gt; entropy = &amp;minus; &amp;Sigma; p&lt;sub&gt;i&lt;/sub&gt; &amp;times;  log&lt;sub&gt;2&lt;/sub&gt; p&lt;sub&gt;i&lt;/sub&gt; &lt;br&gt; This is at a maximum when all the p&lt;sub&gt;i&lt;/sub&gt; are equal, which slightly paradoxically can be seen as a state of maximum disorder, in the sense that therre is not a preference to one value or another.  The term originated in the study of thermodyyamincs in physics, but was coined as an information measure by Shannon and Weaver in the late 1940s
        )

    [entropy measure] =&gt; Array
        (
            [slug] =&gt; entropy measure
            [parent] =&gt; 
            [name] =&gt; entropy measure
            [canonical] =&gt; entropy
            [see] =&gt; 
            [description] =&gt; 
        )

    [epiphenomenon] =&gt; Array
        (
            [slug] =&gt; epiphenomenon
            [parent] =&gt; 
            [name] =&gt; epiphenomenon
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; An epiphenomenon is a secondary effect.  Within the {{philosophy of consciousness}}, it refers to the notion that consciousness does not have a function as such, but has arisen (or its appearance has arisen), as an accident of other processes.
        )

    [epistemic action] =&gt; Array
        (
            [slug] =&gt; epistemic action
            [parent] =&gt; 
            [name] =&gt; epistemic action
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; An epistemic action is an action the purpose of which is to gain information.  Low level examples include turning one's head to look at something you glimpses from the corner of your eye,  or openinga. box to see what's inside.  At a higher level you might look something up on the web, or go to a museum.  This is important in {{comouter vision}} as many {{ambiguous images}} resolve quickly if the veiwpoint is changed even slightly by changing camera position, a process known as {{actuve vision}}. 
        )

    [epistemic interaction] =&gt; Array
        (
            [slug] =&gt; epistemic interaction
            [parent] =&gt; 
            [name] =&gt; epistemic interaction
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; Epistemic interaction refers to the way that {{intelligent user interfaces}} can be designed so that user interactions reveal more information about the users preferences or goals and thus improve future interactions.  For example, if the user hovers over a glossary entry in the text it shows they have some interest in the topic, whereas if the entries are permanently shown in a sidebar it is impossible to know which the user has read.  The former may be prefered as it provides better knowledge, even if marginally less usable.
        )

    [equilibrium] =&gt; Array
        (
            [slug] =&gt; equilibrium
            [parent] =&gt; 
            [name] =&gt; equilibrium
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; 
        )

    [euclidean distance] =&gt; Array
        (
            [slug] =&gt; euclidean distance
            [parent] =&gt; 
            [name] =&gt; Euclidean distance
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; Euclidean distance is the straight-line 'as the crows flies' distance between two points.  In two dimensions it can be calculated uisng Pythagoras theorem and the equivalent for higher dimensions.  Alternative distance measures include the {{Manhatten block distance}}, which is the sum of the x and y distances.
        )

    [european general data protection regulation] =&gt; Array
        (
            [slug] =&gt; european general data protection regulation
            [parent] =&gt; 
            [name] =&gt; European General Data Protection Regulation
            [canonical] =&gt; 
            [see] =&gt; gdpr
            [description] =&gt; 
        )

    [event] =&gt; Array
        (
            [slug] =&gt; event
            [parent] =&gt; 
            [name] =&gt; event
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; 
        )

    [evidence-based medicine] =&gt; Array
        (
            [slug] =&gt; evidence-based medicine
            [parent] =&gt; 
            [name] =&gt; evidence-based medicine
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; Evidence-based medicine is when clinical advice or decsions are based on established empirical studies or other forms of scientific evidence.  At first it seems as though all modern medicine would fall under this category, but there are many areas, for example in psychiatry, where the level of individual difference and complexity of problems makes this kind of evidence hard to gather, or, it may be argued, inappropriate.  Given the range of conditions and symptoms, it is impossible for a doctor or other medical practitioner to keep up to date or remember all the evdience that is published.  This is one of the ways that {{expert systems}} can be an invaluable part of medical practice.
        )

    [exception sets] =&gt; Array
        (
            [slug] =&gt; exception sets
            [parent] =&gt; 
            [name] =&gt; exception sets
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; When dealing with large data sets, for example, during {{data preparation}} of {{big data}}, one usually attempts to apply generic coded rules to every data item.  However, it can sometimes be hard to craete rules that aply to everything, for example, to deal with &quot;Henry VIII&quot; or &quot;The Artist Formerly Known as Prince&quot; in ruels fpr processing names.  Rather than creating ever more complex, and potentially buggy, rules; one can instead create &lt;em&gt;exception sets&lt;/em&gt; that say, for example, that &quot;Henry VIII&quot; is a valid name, or &quot;treat the age field in record 7932 as 88&quot;.  These exception sets are then applied before generic rules as a form of {{override}}.  This is especially important when the input data set may beuodated and then re-processed.
        )

    [exclusive or problem] =&gt; Array
        (
            [slug] =&gt; exclusive or problem
            [parent] =&gt; 
            [name] =&gt; exclusive or problem
            [canonical] =&gt; 
            [see] =&gt; xor problem
            [description] =&gt; 
        )

    [expert knowledge] =&gt; Array
        (
            [slug] =&gt; expert knowledge
            [parent] =&gt; 
            [name] =&gt; expert knowledge
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; 
        )

    [expert system] =&gt; Array
        (
            [slug] =&gt; expert system
            [parent] =&gt; 
            [name] =&gt; expert system
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; 
            [subentries] =&gt; Array
                (
                    [expert system!brittleness] =&gt; Array
                        (
                            [slug] =&gt; expert system!brittleness
                            [parent] =&gt; expert system
                            [name] =&gt; brittleness
                            [canonical] =&gt; 
                            [see] =&gt; 
                            [description] =&gt; 
                        )

                    [expert system!hybrid] =&gt; Array
                        (
                            [slug] =&gt; expert system!hybrid
                            [parent] =&gt; expert system
                            [name] =&gt; hybrid
                            [canonical] =&gt; 
                            [see] =&gt; 
                            [description] =&gt; 
                        )

                    [expert system!limitations] =&gt; Array
                        (
                            [slug] =&gt; expert system!limitations
                            [parent] =&gt; expert system
                            [name] =&gt; limitations
                            [canonical] =&gt; 
                            [see] =&gt; 
                            [description] =&gt; 
                        )

                    [expert system!meta-knowledge] =&gt; Array
                        (
                            [slug] =&gt; expert system!meta-knowledge
                            [parent] =&gt; expert system
                            [name] =&gt; meta-knowledge
                            [canonical] =&gt; 
                            [see] =&gt; 
                            [description] =&gt; 
                        )

                    [expert system!purpose] =&gt; Array
                        (
                            [slug] =&gt; expert system!purpose
                            [parent] =&gt; expert system
                            [name] =&gt; purpose
                            [canonical] =&gt; 
                            [see] =&gt; 
                            [description] =&gt; 
                        )

                    [expert system!rule tracing] =&gt; Array
                        (
                            [slug] =&gt; expert system!rule tracing
                            [parent] =&gt; expert system
                            [name] =&gt; rule tracing
                            [canonical] =&gt; 
                            [see] =&gt; 
                            [description] =&gt; 
                        )

                    [expert system!shell] =&gt; Array
                        (
                            [slug] =&gt; expert system!shell
                            [parent] =&gt; expert system
                            [name] =&gt; shell
                            [canonical] =&gt; 
                            [see] =&gt; 
                            [description] =&gt; 
                        )

                    [expert system!verification] =&gt; Array
                        (
                            [slug] =&gt; expert system!verification
                            [parent] =&gt; expert system
                            [name] =&gt; verification
                            [canonical] =&gt; 
                            [see] =&gt; 
                            [description] =&gt; 
                        )

                )

        )

    [expert system shell] =&gt; Array
        (
            [slug] =&gt; expert system shell
            [parent] =&gt; 
            [name] =&gt; expert system shell
            [canonical] =&gt; expert system!shell
            [see] =&gt; 
            [description] =&gt; 
        )

    [expert system!brittleness] =&gt; Array
        (
            [slug] =&gt; expert system!brittleness
            [parent] =&gt; expert system
            [name] =&gt; brittleness
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; 
        )

    [expert system!hybrid] =&gt; Array
        (
            [slug] =&gt; expert system!hybrid
            [parent] =&gt; expert system
            [name] =&gt; hybrid
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; 
        )

    [expert system!limitations] =&gt; Array
        (
            [slug] =&gt; expert system!limitations
            [parent] =&gt; expert system
            [name] =&gt; limitations
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; 
        )

    [expert system!meta-knowledge] =&gt; Array
        (
            [slug] =&gt; expert system!meta-knowledge
            [parent] =&gt; expert system
            [name] =&gt; meta-knowledge
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; 
        )

    [expert system!purpose] =&gt; Array
        (
            [slug] =&gt; expert system!purpose
            [parent] =&gt; expert system
            [name] =&gt; purpose
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; 
            [subentries] =&gt; Array
                (
                    [expert system!purpose!current state] =&gt; Array
                        (
                            [slug] =&gt; expert system!purpose!current state
                            [parent] =&gt; expert system!purpose
                            [name] =&gt; current state
                            [canonical] =&gt; 
                            [see] =&gt; 
                            [description] =&gt; 
                        )

                    [expert system!purpose!prediction] =&gt; Array
                        (
                            [slug] =&gt; expert system!purpose!prediction
                            [parent] =&gt; expert system!purpose
                            [name] =&gt; prediction
                            [canonical] =&gt; 
                            [see] =&gt; 
                            [description] =&gt; 
                        )

                )

        )

    [expert system!purpose!current state] =&gt; Array
        (
            [slug] =&gt; expert system!purpose!current state
            [parent] =&gt; expert system!purpose
            [name] =&gt; current state
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; 
        )

    [expert system!purpose!prediction] =&gt; Array
        (
            [slug] =&gt; expert system!purpose!prediction
            [parent] =&gt; expert system!purpose
            [name] =&gt; prediction
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; 
        )

    [expert system!rule tracing] =&gt; Array
        (
            [slug] =&gt; expert system!rule tracing
            [parent] =&gt; expert system
            [name] =&gt; rule tracing
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; 
        )

    [expert system!shell] =&gt; Array
        (
            [slug] =&gt; expert system!shell
            [parent] =&gt; expert system
            [name] =&gt; shell
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; 
        )

    [expert system!verification] =&gt; Array
        (
            [slug] =&gt; expert system!verification
            [parent] =&gt; expert system
            [name] =&gt; verification
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; 
        )

    [expert systems] =&gt; Array
        (
            [slug] =&gt; expert systems
            [parent] =&gt; 
            [name] =&gt; expert systems
            [canonical] =&gt; expert system
            [see] =&gt; 
            [description] =&gt; 
        )

    [expert systems shells] =&gt; Array
        (
            [slug] =&gt; expert systems shells
            [parent] =&gt; 
            [name] =&gt; expert systems shells
            [canonical] =&gt; expert system!shell
            [see] =&gt; 
            [description] =&gt; 
        )

    [explainable] =&gt; Array
        (
            [slug] =&gt; explainable
            [parent] =&gt; 
            [name] =&gt; explainable
            [canonical] =&gt; explainable ai
            [see] =&gt; 
            [description] =&gt; 
        )

    [explainable ai] =&gt; Array
        (
            [slug] =&gt; explainable ai
            [parent] =&gt; 
            [name] =&gt; explainable AI
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; 
        )

    [explanation component] =&gt; Array
        (
            [slug] =&gt; explanation component
            [parent] =&gt; 
            [name] =&gt; explanation component
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; 
        )

    [explanation-based learning] =&gt; Array
        (
            [slug] =&gt; explanation-based learning
            [parent] =&gt; 
            [name] =&gt; explanation-based learning
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; 
        )

    [explicit confirmation] =&gt; Array
        (
            [slug] =&gt; explicit confirmation
            [parent] =&gt; 
            [name] =&gt; explicit confirmation
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; 
        )

    [exploitation] =&gt; Array
        (
            [slug] =&gt; exploitation
            [parent] =&gt; 
            [name] =&gt; exploitation
            [canonical] =&gt; exploration-exploitation trade-off
            [see] =&gt; 
            [description] =&gt; 
        )

    [exploration-exploitation trade-off] =&gt; Array
        (
            [slug] =&gt; exploration-exploitation trade-off
            [parent] =&gt; 
            [name] =&gt; exploration-exploitation trade-off
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; 
        )

    [exploration] =&gt; Array
        (
            [slug] =&gt; exploration
            [parent] =&gt; 
            [name] =&gt; exploration
            [canonical] =&gt; exploration-exploitation trade-off
            [see] =&gt; 
            [description] =&gt; 
        )

    [exponential decay] =&gt; Array
        (
            [slug] =&gt; exponential decay
            [parent] =&gt; 
            [name] =&gt; exponential decay
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; Exponential decay is when a numerical value reduces by a fixed proportion in each unit of time (or some other variable).  For example, if the amount of water reduces by a third every day. When this happens the value at time t is K&lt;sub&gt;t&lt;/sup&gt; where K is a constant less than 1, or alternatively e&lt;sup&gt;&ndash;&amp;lambda;t&lt;/sup&gt; where  &amp;lambda; = ln(K), the {{natural logarithm}} of K.  Exponential decay reduces faster than any positive power of 1/t.
        )

    [exponential growth] =&gt; Array
        (
            [slug] =&gt; exponential growth
            [parent] =&gt; 
            [name] =&gt; exponential growth
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; Exponential growth is when a numerical value increases by a fixed proportion in each unit of time (or some other variable).  For example, if the value of gold grows by 10% each year.  When this happens the value at time t is K&lt;sub&gt;t&lt;/sup&gt; where K is a constant greater than 1, or alternatively e&lt;sup&gt;&amp;lambda;t&lt;/sup&gt; where  &amp;lambda; = ln(K), the {{natural logarithm}} of K.  Exponential growth increases faster than any positive power of t.
        )

    [extensional] =&gt; Array
        (
            [slug] =&gt; extensional
            [parent] =&gt; 
            [name] =&gt; extensional
            [canonical] =&gt; extensional representation
            [see] =&gt; 
            [description] =&gt; 
        )

    [extensional representation] =&gt; Array
        (
            [slug] =&gt; extensional representation
            [parent] =&gt; 
            [name] =&gt; extensional representation
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; For a {{logical expression}} or {{database query}} the extensional representation is the collection of all values that satisfy the expression or are returned by the query.  This is in contrast to the {{intensional representation}}, which is the formulae of the expression or the query expresson in a notation such as {{SQL}}.
        )

    [f score] =&gt; Array
        (
            [slug] =&gt; f score
            [parent] =&gt; 
            [name] =&gt; F score
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; The F score is an {{accuracy}} metric that combines {{precision}} and {{recall}}.  It is defined as the harmonic mean of {{precision}} and {{recall}}, that is: &lt;br&gt;1/F = 0.5*( 1/precision + 1/recall ) &lt;br&gt;. There are variants (F&lt;sub&gt;&amp;beta;&lt;/sub&gt; that weight {{precision}} and {{recall}} differently.  Also not to be confused with the {{F distribution}} used in statistical testing.
        )

    [facial recognition] =&gt; Array
        (
            [slug] =&gt; facial recognition
            [parent] =&gt; 
            [name] =&gt; facial recognition
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; Facial recognition refers to {{image processing}} systems thart are able to identify faces in images or video and then identify who the person is, using some form of data bases of faces.  This can be very useful, for exampel, in naming family members on a personal photo album, or preventing a terrorust attack.  However, there are also worries about issues of {{privacy}} and potential misue by authoritarian governments.  These worries have beeb made more severe guven that many facial recognition systems exhibit leels of {{bias}} often being less effective for darker skins and hence more likely to erroneoulsy flag a black person as a suspected criminal.
        )

    [fast fourier transform] =&gt; Array
        (
            [slug] =&gt; fast fourier transform
            [parent] =&gt; 
            [name] =&gt; Fast Fourier Transform
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; A Fast Fourier Transform (FFT) is, as the name suggests, a fast way to perform a {{Fourier transform}} of {{time-series}} or other sequential data.  It exploits symetries in the cacklulatio of the {{Fourier transform}} in order to reuse intermeidate results.  Whereas a brute force version of {{Fourier transform}} is quandratic in the size of gthe input (N&lt;sup&gt;2&lt;/sup&gt;), the FFT is log-linear (N log&lt;sub&gt;2&lt;/sub&gt;N).  For example, for N=1024, FFT is around 100 times faster (1024/log&lt;sub&gt;2&lt;/sub&gt;(1024)).
        )

    [fault-tolerant] =&gt; Array
        (
            [slug] =&gt; fault-tolerant
            [parent] =&gt; 
            [name] =&gt; fault-tolerant
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; 
        )

    [feasibility] =&gt; Array
        (
            [slug] =&gt; feasibility
            [parent] =&gt; 
            [name] =&gt; feasibility
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; 
        )

    [feasible] =&gt; Array
        (
            [slug] =&gt; feasible
            [parent] =&gt; 
            [name] =&gt; feasible
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; 
        )

    [feature interaction] =&gt; Array
        (
            [slug] =&gt; feature interaction
            [parent] =&gt; 
            [name] =&gt; feature interaction
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; 
        )

    [feature sets] =&gt; Array
        (
            [slug] =&gt; feature sets
            [parent] =&gt; 
            [name] =&gt; feature sets
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; 
        )

    [federated learning] =&gt; Array
        (
            [slug] =&gt; federated learning
            [parent] =&gt; 
            [name] =&gt; federated learning
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; 
        )

    [feedback] =&gt; Array
        (
            [slug] =&gt; feedback
            [parent] =&gt; 
            [name] =&gt; feedback
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; 
        )

    [feedback effects] =&gt; Array
        (
            [slug] =&gt; feedback effects
            [parent] =&gt; 
            [name] =&gt; feedback effects
            [canonical] =&gt; feedback
            [see] =&gt; 
            [description] =&gt; 
        )

    [feeling] =&gt; Array
        (
            [slug] =&gt; feeling
            [parent] =&gt; 
            [name] =&gt; feeling
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; In {{psychology}} {{emotion}} refers to physiological effects such heart rate, whereas &lt;em&gt;feeling&lt;/em&gt; refers to the interpretation of this in higher-order ways such as 'fear' or 'excitement'.  Crucially in many situations the physiological effect ({{emotion}}) preceeds the interpretation ({{feeling}}).  For example,  if soemone burst a aballoon your heart will instantly race, butwhether this leads to a sense of fun, fear or anger depends on the situation.   In day-to-day speech both emotion and feeling are often called 'emotion'.  The distinction is important, however, when creating {{intelligent user interfaces}} that seek to interpret or predict human emotional responses.  It can also be useful when considering {{artificial emotion}}, with different ways to emulate the physiological side and the interpretation.
        )

    [fft] =&gt; Array
        (
            [slug] =&gt; fft
            [parent] =&gt; 
            [name] =&gt; FFT
            [canonical] =&gt; fast fourier transform
            [see] =&gt; 
            [description] =&gt; 
        )

    [figure-ground separation] =&gt; Array
        (
            [slug] =&gt; figure-ground separation
            [parent] =&gt; 
            [name] =&gt; figure-ground separation
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; In a visual scene or image, figure-ground separation is the process of distinguising the background, for example the room or landscape (&lt;em&gt;ground&lt;/em&gt;) from things in the foreground, such as a person or car (the &lt;em&gt;figure&lt;/em&gt;).  Figure-ground separation is often performed at an early stae in an image-processing pipeline before more complex processing to identify the figure.
        )

    [filter bubbles] =&gt; Array
        (
            [slug] =&gt; filter bubbles
            [parent] =&gt; 
            [name] =&gt; filter bubbles
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; 
        )

    [finite impulse response] =&gt; Array
        (
            [slug] =&gt; finite impulse response
            [parent] =&gt; 
            [name] =&gt; finite impulse response
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; In a finite input response (FIR) system observations or events far enough in the past have no effect on the current behaviour of the system.  Typically this is true of systems that use a fixed {{window}} ovef the data of, say, the last N items and base responses only on these.  However, this may also be less obvious.  Some statistical time series analysis techniques ar FIR such as  a {{moving average}} as well as {[window-based}} neural networks.
        )

    [finite state machine] =&gt; Array
        (
            [slug] =&gt; finite state machine
            [parent] =&gt; 
            [name] =&gt; finite state machine
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; 
        )

    [fir] =&gt; Array
        (
            [slug] =&gt; fir
            [parent] =&gt; 
            [name] =&gt; FIR
            [canonical] =&gt; finite impulse response
            [see] =&gt; 
            [description] =&gt; 
        )

    [firewalls] =&gt; Array
        (
            [slug] =&gt; firewalls
            [parent] =&gt; 
            [name] =&gt; firewalls
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; 
        )

    [first-order difference] =&gt; Array
        (
            [slug] =&gt; first-order difference
            [parent] =&gt; 
            [name] =&gt; first-order difference
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; 
        )

    [first-order predicate calculus] =&gt; Array
        (
            [slug] =&gt; first-order predicate calculus
            [parent] =&gt; 
            [name] =&gt; first-order predicate calculus
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; 
        )

    [fitness function] =&gt; Array
        (
            [slug] =&gt; fitness function
            [parent] =&gt; 
            [name] =&gt; fitness function
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; In {{optimisation}} and forms of {{machine learniong}} the fitness function (also known as objective function or utility function), measures how good a solution or state is.  For example, the fitness function for an image reconstruction may be measured as the sum of the squares of difference between the pixel values of the target and generated images.  Note that the choice of fitness function has important impact on efficiency, for example, if it has hard {{thresholds}} this may make incemental learning hard.  The fitness function can also be a source of {{bias}}, for example, if you choose based on popularity you may effectively embody societal discrimination.
        )

    [fitness landscape] =&gt; Array
        (
            [slug] =&gt; fitness landscape
            [parent] =&gt; 
            [name] =&gt; fitness landscape
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; 
        )

    [fixation point] =&gt; Array
        (
            [slug] =&gt; fixation point
            [parent] =&gt; 
            [name] =&gt; fixation point
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; In a camera, the fixation point is where the camera is 'looking'.  That is the point in the world that is at the centre of the camera image.  
        )

    [flocking behaviour] =&gt; Array
        (
            [slug] =&gt; flocking behaviour
            [parent] =&gt; 
            [name] =&gt; flocking behaviour
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; Many animals including birds and sheep exhibit flocking behaviour; that is they appear to move as an almost fluid whole.  In Starlings this is called a murmuration, in fish shoaling, and for many insects swarming.  The {{emergent behaiour}} arises because of many local interactions, each bird, animal or insect is aware of its neighbours and seeks to avoid bumping into them whikst alsos taying close; some have an idea ow ehere they woudl like to go or avoid, perhaps becasue they have spotted a good food source or a preditor.  These many small interactions lead to the overall flocking behaviour.  This can be modelled mathematically for biologists to understand the animals and also to be used in {{CGI}} for films.  These behaviours can also be used as an inspiration for computational {{swarm algorithms}}, which use large numbers of indivdiually simple {{computational agents}} to tackle complex problems.
        )

    [fmri] =&gt; Array
        (
            [slug] =&gt; fmri
            [parent] =&gt; 
            [name] =&gt; fMRI
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; Functional magnetic resonance imaging (fMRI) is a form of brain scanning that uses levels of blood oxygen to determine which parts of the brain are most active at any moment.  This is used as part of medical diagnoses, but also by neuroscientists to map the interconnections between parts of the brain.  This data is in turn being used by AI researchers, incuding the {{Human Brain Project}}, to create artifical models of human brain functioning and to inspire more human-like {{neural network}] algorithms.
        )

    [focus] =&gt; Array
        (
            [slug] =&gt; focus
            [parent] =&gt; 
            [name] =&gt; focus
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; In a camera, the focus is the the distance at which objects are most sharp.  It is usually controlled by adjusting the dustance between the lens and the light detector.
        )

    [forgetful hill climbing] =&gt; Array
        (
            [slug] =&gt; forgetful hill climbing
            [parent] =&gt; 
            [name] =&gt; forgetful hill climbing
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; The simplest form of {{hill climbing}} search is 'forgetful', that is at each step one moves to an adjoining or close point  with a better (or best) {{fitness function}}.  The is has the dabger of getting stuck as a {{local maxima}} or reach an impasse if some states are not allowed, for example those thart are not {{feasible}} when there are {[constraints}}.  For this reason, variants of hill climbing may searches keep track of past states (memory) and use {{backtracking}} to reisit them in case of impasse, or to explore alternatives suhc as anext best solution.
        )

    [forward branching factor] =&gt; Array
        (
            [slug] =&gt; forward branching factor
            [parent] =&gt; 
            [name] =&gt; forward branching factor
            [canonical] =&gt; 
            [see] =&gt; branching factor
            [description] =&gt; 
        )

    [forward chaining] =&gt; Array
        (
            [slug] =&gt; forward chaining
            [parent] =&gt; 
            [name] =&gt; forward chaining
            [canonical] =&gt; 
            [see] =&gt; forward reasoning
            [description] =&gt; 
        )

    [forwards chaining] =&gt; Array
        (
            [slug] =&gt; forwards chaining
            [parent] =&gt; 
            [name] =&gt; forwards chaining
            [canonical] =&gt; forward chaining
            [see] =&gt; 
            [description] =&gt; 
        )

    [forward reasoning] =&gt; Array
        (
            [slug] =&gt; forward reasoning
            [parent] =&gt; 
            [name] =&gt; forward reasoning
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; Forward reasoning (also known as data-driven reasoning) is used in {{reasoning systems}} when rules are run 'forwards' in order to work out further things form the knowledge one has.  For example, given &quot;Scrates is a man&quot; and &quot;all men are mortal&quot;, one can work out that &quot;Socrates is a man&quot;, or ina. numerical example, given &quot;x&gt;y&quot;, &quot;y&gt;z&quot;, and &quot;z=3&quot; one can work out in one step that that &quot;x&gt;z&quot; and &quot;y&gt;3&quot; and then in a second reasoning step (through towo paths) that &quot;x&gt;3&quot;.    Forward reasoning is particularly useful when the forwards {{branching factor}} is smaller than the backwards {{branching factor}}, for example when there is a relatively small numbers of known facts and rules to combine, or when there is a need to precomputer derived knwoledge as a pre-computation step in order to speed up reasoning after deployment.
        )

    [foundation models] =&gt; Array
        (
            [slug] =&gt; foundation models
            [parent] =&gt; 
            [name] =&gt; foundation models
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; 
        )

    [four knights problem] =&gt; Array
        (
            [slug] =&gt; four knights problem
            [parent] =&gt; 
            [name] =&gt; four knights problem
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; 
        )

    [fourier analysis] =&gt; Array
        (
            [slug] =&gt; fourier analysis
            [parent] =&gt; 
            [name] =&gt; fourier analysis
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; 
        )

    [fourier transform] =&gt; Array
        (
            [slug] =&gt; fourier transform
            [parent] =&gt; 
            [name] =&gt; Fourier transform
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; A Fourier transform converts {{time series}} or other sequential data into a a combination of sine and cosine waves of different frequencies.  This is similar to the frequency level indicators that are seen in high-end amplifiers, but at a much finer scale.  Typically the signal is split into frequencies using some power of 2, as this can be calculated using a {{Fast Fourier Transform}}, but in principle any range of consecutive frequencies can be used.
        )

    [fourier transforms] =&gt; Array
        (
            [slug] =&gt; fourier transforms
            [parent] =&gt; 
            [name] =&gt; fourier transforms
            [canonical] =&gt; fourier transform
            [see] =&gt; 
            [description] =&gt; 
        )

    [fovea] =&gt; Array
        (
            [slug] =&gt; fovea
            [parent] =&gt; 
            [name] =&gt; fovea
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; The fovea is the central portion of retina on the back of the eye.  The retina consists of light sensing cells, but they are not uniformly spaced, in fact the majority are in a small area towards the centre of vision, the fovea.  In practice this means that a very small region of your vision, roughly the size of your thumb at arms length, is seen at high resolution with increasingly less resolution as you move away from it.  Your impression of a complete detailed view is constructed by your brain in part as your eye rapidly jumps across the scene (known as {{saccades}}) and in part based on memory and past experience.
        )

    [fpgas] =&gt; Array
        (
            [slug] =&gt; fpgas
            [parent] =&gt; 
            [name] =&gt; FPGAs
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; FPGAs (field programable gate arrays) are chips comprised of simple logic functions such as AND/OR/NOT interlinked with 'gates' that can be modified, rather link in a memory chip. By chosing which of these gates are open and which closed, a FPGA can be programmed to perform patterns of logic operations almost aas fast as if they had been hard coded into the logic.  
        )

    [frame] =&gt; Array
        (
            [slug] =&gt; frame
            [parent] =&gt; 
            [name] =&gt; frame
            [canonical] =&gt; frame problem
            [see] =&gt; 
            [description] =&gt; 
        )

    [frames] =&gt; Array
        (
            [slug] =&gt; frames
            [parent] =&gt; 
            [name] =&gt; frames
            [canonical] =&gt; frame problem
            [see] =&gt; 
            [description] =&gt; 
        )

    [frame axiom] =&gt; Array
        (
            [slug] =&gt; frame axiom
            [parent] =&gt; 
            [name] =&gt; frame axiom
            [canonical] =&gt; frame problem
            [see] =&gt; 
            [description] =&gt; 
        )

    [frame axioms] =&gt; Array
        (
            [slug] =&gt; frame axioms
            [parent] =&gt; 
            [name] =&gt; frame axioms
            [canonical] =&gt; frame problem
            [see] =&gt; 
            [description] =&gt; 
        )

    [frame problem] =&gt; Array
        (
            [slug] =&gt; frame problem
            [parent] =&gt; 
            [name] =&gt; frame problem
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; 
        )

    [frankenstein] =&gt; Array
        (
            [slug] =&gt; frankenstein
            [parent] =&gt; 
            [name] =&gt; Frankenstein
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; Frankenstein is a book written by {{Mary Shelly}} in 1818.  The name is often used to refer to the Hammer Movies monster, but strictly it is the name of the monster's creator, Victor Frankenstein.  It is story of scientific hubris.  Frankenstein used a combination of body parts animated by electricty to create a new living being.  Frankenstein's initial excitement turns to disgust and fear.  It is often used as a warning as to the dangers of {{robotics}} and AI.  While the monster had no mechanical or electronic parts, so not even a {{cyborg}}, the idea of creating an artificial lifeform has many parallels with {{strong AI}} in particular.
        )

    [frequency domain] =&gt; Array
        (
            [slug] =&gt; frequency domain
            [parent] =&gt; 
            [name] =&gt; frequency domain
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; 
        )

    [frequency space] =&gt; Array
        (
            [slug] =&gt; frequency space
            [parent] =&gt; 
            [name] =&gt; frequency space
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; 
        )

    [fully connected] =&gt; Array
        (
            [slug] =&gt; fully connected
            [parent] =&gt; 
            [name] =&gt; fully connected
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; 
        )

    [function allocation] =&gt; Array
        (
            [slug] =&gt; function allocation
            [parent] =&gt; 
            [name] =&gt; function allocation
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; 
        )

    [fuzzy logic] =&gt; Array
        (
            [slug] =&gt; fuzzy logic
            [parent] =&gt; 
            [name] =&gt; fuzzy logic
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; 
        )

    [fuzzy reasoning] =&gt; Array
        (
            [slug] =&gt; fuzzy reasoning
            [parent] =&gt; 
            [name] =&gt; fuzzy reasoning
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; 
        )

    [fuzzy set theory] =&gt; Array
        (
            [slug] =&gt; fuzzy set theory
            [parent] =&gt; 
            [name] =&gt; fuzzy set theory
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; 
        )

    [fuzzy sets] =&gt; Array
        (
            [slug] =&gt; fuzzy sets
            [parent] =&gt; 
            [name] =&gt; fuzzy sets
            [canonical] =&gt; fuzzy set theory
            [see] =&gt; 
            [description] =&gt; 
        )

    [game of life] =&gt; Array
        (
            [slug] =&gt; game of life
            [parent] =&gt; 
            [name] =&gt; Game of Life
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; Game of Life is a simple two-state {{cellular automata}} based on a rectangular grid of squares, where each square is colured as active (alive) or a background color (dead or empty).  If an empty cell has three live neighbours it becomes active/alive.  If a live cell has two many live neighbours (4 or more) it is overcrowded and dies.  If it has only one live neighbour it is lonely and also dies.  From these simple rules many dynamic patterns emerge.
        )

    [game playing] =&gt; Array
        (
            [slug] =&gt; game playing
            [parent] =&gt; 
            [name] =&gt; game playing
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; 
            [subentries] =&gt; Array
                (
                    [game playing!heuristics] =&gt; Array
                        (
                            [slug] =&gt; game playing!heuristics
                            [parent] =&gt; game playing
                            [name] =&gt; heuristics
                            [canonical] =&gt; 
                            [see] =&gt; 
                            [description] =&gt; 
                        )

                )

        )

    [game playing!heuristics] =&gt; Array
        (
            [slug] =&gt; game playing!heuristics
            [parent] =&gt; game playing
            [name] =&gt; heuristics
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; 
        )

    [game theory] =&gt; Array
        (
            [slug] =&gt; game theory
            [parent] =&gt; 
            [name] =&gt; game theory
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; Game theory is the mathematical study of games.  A 'game' is defined quite formally and is not about fun or enjoyment, but more about rewards and penalties of game play.
        )

    [game tree] =&gt; Array
        (
            [slug] =&gt; game tree
            [parent] =&gt; 
            [name] =&gt; game tree
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; A game tree is a tree representation of game states and moves.  Nodes in the tree are game states and branches are labelled by the player who made them.  In two-person, turn-taking games the player alternates at each level of the tree.  A particular path through the tree represents a particular playing of the game. 
        )

    [game trees] =&gt; Array
        (
            [slug] =&gt; game trees
            [parent] =&gt; 
            [name] =&gt; game trees
            [canonical] =&gt; game tree
            [see] =&gt; 
            [description] =&gt; 
        )

    [gan] =&gt; Array
        (
            [slug] =&gt; gan
            [parent] =&gt; 
            [name] =&gt; GAN
            [canonical] =&gt; 
            [see] =&gt; generative adversarial network
            [description] =&gt; 
        )

    [garry kasparov] =&gt; Array
        (
            [slug] =&gt; garry kasparov
            [parent] =&gt; 
            [name] =&gt; garry kasparov
            [canonical] =&gt; kasparov, garry
            [see] =&gt; 
            [description] =&gt; 
        )

    [gaussian distribution] =&gt; Array
        (
            [slug] =&gt; gaussian distribution
            [parent] =&gt; 
            [name] =&gt; Gaussian distribution
            [canonical] =&gt; 
            [see] =&gt; normal distribution
            [description] =&gt; 
        )

    [gaussian filter] =&gt; Array
        (
            [slug] =&gt; gaussian filter
            [parent] =&gt; 
            [name] =&gt; Gaussian filter
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; 
        )

    [gaussian filters] =&gt; Array
        (
            [slug] =&gt; gaussian filters
            [parent] =&gt; 
            [name] =&gt; gaussian filters
            [canonical] =&gt; gaussian filter
            [see] =&gt; 
            [description] =&gt; 
        )

    [gdpr] =&gt; Array
        (
            [slug] =&gt; gdpr
            [parent] =&gt; 
            [name] =&gt; GDPR
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; General Data Protection Regulation (GDPR) is the European Union data protection law 
        )

    [general knowledge] =&gt; Array
        (
            [slug] =&gt; general knowledge
            [parent] =&gt; 
            [name] =&gt; general knowledge
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; The technical meaning of 'general knowledge' is just what it says, the kinds of knowledge about the world that we all have: the sky is blue, apples fall down, people generally have two arms and two legs.  This is contrasted with more specialised knowledge about a specifi topic (such as quantum mechanics, or 15th Century Kings of France) or a specific situation (such as the locations of furniture in your home).  Until recently it has been very difficult for AI to deal with problems involving general knowledge, however, {{large language models}} trained on vast qunatities of human text are beginning the change this as are systems such as {{IBM Watson}} whihc use a combination of rule-based approaches, pre-processed data and real time access to large data sources.
        )

    [general or common-sense knowledge] =&gt; Array
        (
            [slug] =&gt; general or common-sense knowledge
            [parent] =&gt; 
            [name] =&gt; general or common-sense knowledge
            [canonical] =&gt; general knowledge
            [see] =&gt; 
            [description] =&gt; 
        )

    [generalisation] =&gt; Array
        (
            [slug] =&gt; generalisation
            [parent] =&gt; 
            [name] =&gt; generalisation
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; Generalisation is the way in which systems take specific examples and construct more abstract rules that cover multiple cases.  In simple {{rote learniong}} one takes examples and simply recalls the expected output or repsonse for each example.   When the same example is given the learner returns a correct result, but is unable to deal with any input that wasn't in the {{training set}}.  Techniques for generalisation attempt to seek more abstract patterns, trends or commonalities in the traing to enable them to respond to novel inputs.   The term also has a specific meaning in the {{ACT-R}} {{cognitive architecture}}.
        )

    [generalization] =&gt; Array
        (
            [slug] =&gt; generalization
            [parent] =&gt; 
            [name] =&gt; generalization
            [canonical] =&gt; generalisation
            [see] =&gt; 
            [description] =&gt; 
        )

    [generate and filter] =&gt; Array
        (
            [slug] =&gt; generate and filter
            [parent] =&gt; 
            [name] =&gt; generate and filter
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; A simple model of {{creativity}} is generate and filter, which can be seen as a human version of {{generate and test}}.  Humans (or AI) create several of potential ideas, more or less good, and then discard those that are less good and pursue further those that are better.  The initial generation process may be entirely random, for example, ink blots, or more guided.  Different kinds of art focus more or less on the initial geberations (e.g. painting) or the critical filterin (e.g. framing a landscape in photography).  Early {{artificial creativity}} often focused on generating good enough options leavung ti to the humanto be the critic and choose the best outputs to display.  {{Generative models}} are improving this generation phase as they are built on vast corpora of past human outputs (of varying quality).  Perhaps the real challenge is for AI to act as the critic, but there are systems that are beginning to tackle this, at east at the level of day-to-day creartivity, such as writing an essay.
        )

    [generate and test] =&gt; Array
        (
            [slug] =&gt; generate and test
            [parent] =&gt; 
            [name] =&gt; generate and test
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; One way to {{search}} for solution to a problem is simply to randomly or systematically &lt;em&gt;generate&lt;/em&gt; a putative solution and then &lt;em&gt;test&lt;/em&gt; whether it solves the problem.  For example, given a game of Sudoku, one can generate squares with randomdigits in and then check if it satisfies the rules of Sudoku (e.g. each digit occurs exactly once in each row and column).  This is usually a highlyinefficient way to {{search}}, as it ignores all of the structure in the {[search space}}.  However, it can be used when the {{search space}} is small to exhaustively check all possibilities. &lt;br&gt;&lt;br&gt; Generate and test can also be used as a startng point to geberate more compelx algi=orithms, for example, by generating {{partial solutions}}, or using {{heuristics}} to guide the generation process.  It can aslo be used theoretically as a comparison point for other algorithms -- if your new complex algorithm is not performing better than generate and test, you know you need to return to the drawing board! 
        )

    [generations] =&gt; Array
        (
            [slug] =&gt; generations
            [parent] =&gt; 
            [name] =&gt; generations
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; In biological populations, different indviduals greo, breed and die ate differnet times, the idea of generation os herev=fore kore an average time from birth to breeding. In a {{genetic algorithm}} the articial process is usually more lock-step, with a population being considered as a group, with {{fitness}} assessment, mutation, breeding and death applied at a single time step.
        )

    [generative] =&gt; Array
        (
            [slug] =&gt; generative
            [parent] =&gt; 
            [name] =&gt; generative
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; 
        )

    [generative adversarial network] =&gt; Array
        (
            [slug] =&gt; generative adversarial network
            [parent] =&gt; 
            [name] =&gt; generative adversarial network
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; 
        )

    [generative adversarial networks] =&gt; Array
        (
            [slug] =&gt; generative adversarial networks
            [parent] =&gt; 
            [name] =&gt; generative adversarial networks
            [canonical] =&gt; generative adversarial network
            [see] =&gt; 
            [description] =&gt; 
        )

    [generative ai] =&gt; Array
        (
            [slug] =&gt; generative ai
            [parent] =&gt; 
            [name] =&gt; generative AI
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; 
        )

    [genes] =&gt; Array
        (
            [slug] =&gt; genes
            [parent] =&gt; 
            [name] =&gt; genes
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; Genes are areas of a {{chromosome}} that form a coding unit, raher like a word or sentance in text.  At a low-level genes are related to the production of various proteins, but at a higher level give rise to inherited traits such as hair colour.  In humans, other animals and many plants, genes come in pairs one inherited from each parent, but there are exceptions, for example, those missing on the (rather stunted) male Y chromosome, or occasionally addtional chromosomes.   The pattern of genes in an animal or plant is called the {{genotype}} whereas the way this appaers (say eye color) is called the {{phenotype}}.  Without {DNA}} testing only the {{phenotype}} is observabe and able to make a difference to the ppearance, behaviour and abilities of the creature or plant. &lt;br&gt;&lt;br&gt;.{{Genetic algorithms}} seek to mimic the way genes are inherited from parents and the survival of prodgeny is dependent on derived features ({{phenotype}}). &lt;br&gt;&lt;br&gt; In early genetics, genes were regarded as having isolated and rule-like behaviour: a blue-eye gene and a brown-eye gene ({{genotype}}) giving rise to a brown eyed person ({{phenotype}}).  However, it is now known that the mapoing is far more coplex, with many genes and environmental factors giving rise to the observable  {{phenotype}}, and also complex interactions within the genome: genes turned on and off, and genes that interact stringly often clusterd togetger on the chromosome so thet they are usually inherited as a unit. &lt;br&gt;&lt;br&gt; This more nuanced understanding of biological genes suggests potential insights for {{genetic algorithms}}, although the majority still are based on 19th century genetics.!
        )

    [genetic algorithm] =&gt; Array
        (
            [slug] =&gt; genetic algorithm
            [parent] =&gt; 
            [name] =&gt; genetic algorithm
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; A genetic algorithm takes inspiration from the proceses of natural selection for biological species.  Indiiduals are defined using a set of paaremeters, effectively artificial {{genes}} -- in the biological analogy, the {{genotype}}.  These are used to assess their {{fitness}} based on whatever criteria are considered optimal and any {{constraints}} -- in the biological analogy, the {{phenotype}}.   The least fit indivdiuals are 'die' (removed from the popukation) and the fitest selected for 'breeding'; typically this process is {{stochastic}}, with fitness being sued to weigh the {{proabability}} of a {{random}} decision as to which die and breed.   Those chosen to reed may be subject to random {{mutation}} and pairs of individuals brought together and theor aetiicial {{genes}} randomly mxed, in a process akin to mating.  In some versions only mutation is used so that the procsses is more like the growth of bacetria or asexual reproduction.  In some varuants the new interbred individuals replace their parents' geeration entirely, in some they are added to it (replacing those that died).
        )

    [genetic algorithms] =&gt; Array
        (
            [slug] =&gt; genetic algorithms
            [parent] =&gt; 
            [name] =&gt; genetic algorithms
            [canonical] =&gt; genetic algorithm
            [see] =&gt; 
            [description] =&gt; 
        )

    [genetic programming] =&gt; Array
        (
            [slug] =&gt; genetic programming
            [parent] =&gt; 
            [name] =&gt; genetic programming
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; Genetic programming is a form of {{genetic algorithm}} where the states of the individual elemnts in a population are some form of program structure.  This is often represented as  a syntax tree, decision tree, or similar structural form.  {{Mutations}} and {{breeding}} are often specially designed to make well formed programs, and the evaluation function usually involves running the programover test cases.
        )

    [genotype] =&gt; Array
        (
            [slug] =&gt; genotype
            [parent] =&gt; 
            [name] =&gt; genotype
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; In biological {{genetics}}  the genotype is the actual gene structure , in contrast to  the {{pheonotype}}, which is the observable characteristics of the organism.  In computational {{genetic algorithms}} there is a similar difference between the way in which the state is encoded, for example, the locations of appliances in a kitchen, and the way that gives rise to, say, plumbing and wiring issues within the kitchen.  Often the mapping between genotype and {{pheonotype}} is effectively a one-way function, where it is hard to work out the geotype given the phenotype.  However, for a {{genetic algorithm}} to work it is essential that there is at least some pattern to the relationship otherwise it will be no better than a {{Monte Carlo algorithm}}.
        )

    [gestalt] =&gt; Array
        (
            [slug] =&gt; gestalt
            [parent] =&gt; 
            [name] =&gt; gestalt
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; Getsalt is a term used in the {{psychology}} of {{perception}} to describe the way that overall features emerge as a whole rather than by piecing together parts.  For example you just see a 'face' rather than inferring it is a face based on the presence of a nose, mouth and eyes.  {{Bottom-up}} {{computer vision}} algorithms often do nto operat like this, but certain kinds of {{neural network}} do appear to emulate aspects of human gestalt.
        )

    [gesture recognition] =&gt; Array
        (
            [slug] =&gt; gesture recognition
            [parent] =&gt; 
            [name] =&gt; gesture recognition
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; 
        )

    [gig economy] =&gt; Array
        (
            [slug] =&gt; gig economy
            [parent] =&gt; 
            [name] =&gt; gig economy
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; The gig economy refers to the internet or app-based business models that grew up in the late 2010s incuding taxi-like services such as Uber and delivery services such as Deliveroo.  They offer flexible working for those providing the services, but often poor job security and working conditions.  
        )

    [gig-economy] =&gt; Array
        (
            [slug] =&gt; gig-economy
            [parent] =&gt; 
            [name] =&gt; gig-economy
            [canonical] =&gt; gig economy
            [see] =&gt; 
            [description] =&gt; 
        )

    [gis] =&gt; Array
        (
            [slug] =&gt; gis
            [parent] =&gt; 
            [name] =&gt; GIS
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; GIS (geographical information systems) refers to computer systems, applications and methods for storing and processing geograohical information such as maps, landuse, locations of landmarks, and routes.  They typically need special ways of storing information (such as {{quad trees}} ) as one is often interested in retreiving things that are, for example, close to location or contained within a region not just exect matches. 
        )

    [glial cells] =&gt; Array
        (
            [slug] =&gt; glial cells
            [parent] =&gt; 
            [name] =&gt; glial cells
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; 
        )

    [global maximum] =&gt; Array
        (
            [slug] =&gt; global maximum
            [parent] =&gt; 
            [name] =&gt; global maximum
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; The global maximum is the point in a {[search space}} or {[fitness landsape}} with the highest possible value.   When a high score is good, the global maximum is the {{optimal}} solution.  It is in contrast to a {{local maximum}}, that is a point in the {{search space}} that is larger than any of its neighbours, but not necessarily largest overall.
        )

    [global minimum] =&gt; Array
        (
            [slug] =&gt; global minimum
            [parent] =&gt; 
            [name] =&gt; global minimum
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; The global minimum is the point in a {[search space}} or {[fitness landsape}} with the lowest possible value.   When a low score is good, the global minimum is the {{optimal}} solution.  It is in contrast to a {{local minimum}}, that is a point in the {{search space}} that is smaller than any of its neighbours, but not necessarily smallest overall.
        )

    [global planning] =&gt; Array
        (
            [slug] =&gt; global planning
            [parent] =&gt; 
            [name] =&gt; global planning
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; 
        )

    [global synchronisation of oscillations] =&gt; Array
        (
            [slug] =&gt; global synchronisation of oscillations
            [parent] =&gt; 
            [name] =&gt; global synchronisation of oscillations
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; 
        )

    [global workspace theory] =&gt; Array
        (
            [slug] =&gt; global workspace theory
            [parent] =&gt; 
            [name] =&gt; global workspace theory
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; 
        )

    [go] =&gt; Array
        (
            [slug] =&gt; go
            [parent] =&gt; 
            [name] =&gt; Go
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; Go is a board game played with white and black pieces on a 19x19 board.  the rules are simple, but the stregtegy complex.  It was generally regarded as too difficult for AI as the {{branching factor}} is so large.  In 2016, however, {{AlphaGo}} defeated a Go grandmaster using a mixture of traditional AI {{game search}} and {{deep learning}}.
        )

    [goal state] =&gt; Array
        (
            [slug] =&gt; goal state
            [parent] =&gt; 
            [name] =&gt; goal state
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; 
        )

    [goal states] =&gt; Array
        (
            [slug] =&gt; goal states
            [parent] =&gt; 
            [name] =&gt; goal states
            [canonical] =&gt; goal state
            [see] =&gt; 
            [description] =&gt; 
        )

    [goal-driven reasoning] =&gt; Array
        (
            [slug] =&gt; goal-driven reasoning
            [parent] =&gt; 
            [name] =&gt; goal-driven reasoning
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; 
        )

    [godel] =&gt; Array
        (
            [slug] =&gt; godel
            [parent] =&gt; 
            [name] =&gt; godel
            [canonical] =&gt; godel, kurt
            [see] =&gt; 
            [description] =&gt; 
        )

    [godel, kurt] =&gt; Array
        (
            [slug] =&gt; godel, kurt
            [parent] =&gt; 
            [name] =&gt; G&amp;ouml;del, Kurt
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; G&amp;ouml;del, Kurt is a mathematicain famous for his incompleteness theorem that proved it was impossible to prove everything that is true in mathematics.  There are some tings that are true, but for which there can never be a proof.  That is, there will always be gaps in knowledge that are fundamental, not simply to hard to tackle yet.
        )

    [google] =&gt; Array
        (
            [slug] =&gt; google
            [parent] =&gt; 
            [name] =&gt; Google
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; 
        )

    [gpt-3] =&gt; Array
        (
            [slug] =&gt; gpt-3
            [parent] =&gt; 
            [name] =&gt; GPT-3
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; GPT-3 is a {{large-language model}} developed by {{OpenAI}}, and was influential in powering {{ChatGPT}}, which radically changed expectations as the the capabilities of {{AI}}.  The aparent step-change in capability started a conversaton as to whether any intelligent behaviour could in fact be produced based on sheer volume of data and computation.   GPT-3 can be tuned for different tasks using prompts and is derisgned to be used as a service by other applicatons, this has led to a large growth of applications using GPT-3 as the underlying engine.  However, becasue GPT-3 and its successors are built using lareg volumes of {{web data}} there are concerns for {{privacy}} and {{intellectual property}}, especially as it occasionally could reproduce verbatim portions of input documents.
        )

    [gpt-4] =&gt; Array
        (
            [slug] =&gt; gpt-4
            [parent] =&gt; 
            [name] =&gt; GPT-4
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; GPT-4 is the successor of {{GPT-3}}.  It used an order of magnitued more training data an weights in its {{deep neural network}}.  It furthered speculation as to the potential for true {{artifcial general intelligence}}.  Fears about  {{intelectual property}} cotnued, but the abaility to create apparently convincing argments (albeit sometimes including entirely fictitious evidence -- called {{hallucinating}}) led to angst in education as it added to existing problems of essay mills and plagiarism.
        )

    [gpu] =&gt; Array
        (
            [slug] =&gt; gpu
            [parent] =&gt; 
            [name] =&gt; GPU
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; A GPU (graphics processing unit) is a hardware chip designed to optimise graphics operations such as rendering images in gaming.  GPUs apply {{parallel processing}} to multiple streams of data simultaneously and have proved useful in data-procesisng applications beyond graphics, notably in neural networks and other AI algorihms.
        )

    [graceful degradation] =&gt; Array
        (
            [slug] =&gt; graceful degradation
            [parent] =&gt; 
            [name] =&gt; graceful degradation
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; 
        )

    [gradient descent] =&gt; Array
        (
            [slug] =&gt; gradient descent
            [parent] =&gt; 
            [name] =&gt; gradient descent
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; 
        )

    [gradient filters] =&gt; Array
        (
            [slug] =&gt; gradient filters
            [parent] =&gt; 
            [name] =&gt; gradient filters
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; 
        )

    [gradient operators] =&gt; Array
        (
            [slug] =&gt; gradient operators
            [parent] =&gt; 
            [name] =&gt; gradient operators
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; 
        )

    [grammar] =&gt; Array
        (
            [slug] =&gt; grammar
            [parent] =&gt; 
            [name] =&gt; grammar
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; 
            [subentries] =&gt; Array
                (
                    [grammar!case] =&gt; Array
                        (
                            [slug] =&gt; grammar!case
                            [parent] =&gt; grammar
                            [name] =&gt; case
                            [canonical] =&gt; 
                            [see] =&gt; 
                            [description] =&gt; 
                        )

                    [grammar!context-free] =&gt; Array
                        (
                            [slug] =&gt; grammar!context-free
                            [parent] =&gt; grammar
                            [name] =&gt; context-free
                            [canonical] =&gt; 
                            [see] =&gt; 
                            [description] =&gt; 
                        )

                    [grammar!context-sensitive] =&gt; Array
                        (
                            [slug] =&gt; grammar!context-sensitive
                            [parent] =&gt; grammar
                            [name] =&gt; context-sensitive
                            [canonical] =&gt; 
                            [see] =&gt; 
                            [description] =&gt; 
                        )

                    [grammar!rules] =&gt; Array
                        (
                            [slug] =&gt; grammar!rules
                            [parent] =&gt; grammar
                            [name] =&gt; rules
                            [canonical] =&gt; 
                            [see] =&gt; 
                            [description] =&gt; 
                        )

                    [grammar!semantic] =&gt; Array
                        (
                            [slug] =&gt; grammar!semantic
                            [parent] =&gt; grammar
                            [name] =&gt; semantic
                            [canonical] =&gt; 
                            [see] =&gt; 
                            [description] =&gt; 
                        )

                    [grammar!syntactic] =&gt; Array
                        (
                            [slug] =&gt; grammar!syntactic
                            [parent] =&gt; grammar
                            [name] =&gt; syntactic
                            [canonical] =&gt; 
                            [see] =&gt; 
                            [description] =&gt; 
                        )

                )

        )

    [grammar!case] =&gt; Array
        (
            [slug] =&gt; grammar!case
            [parent] =&gt; grammar
            [name] =&gt; case
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; 
        )

    [grammar!context-free] =&gt; Array
        (
            [slug] =&gt; grammar!context-free
            [parent] =&gt; grammar
            [name] =&gt; context-free
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; 
        )

    [grammar!context-sensitive] =&gt; Array
        (
            [slug] =&gt; grammar!context-sensitive
            [parent] =&gt; grammar
            [name] =&gt; context-sensitive
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; 
        )

    [grammar!rules] =&gt; Array
        (
            [slug] =&gt; grammar!rules
            [parent] =&gt; grammar
            [name] =&gt; rules
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; 
        )

    [grammar!semantic] =&gt; Array
        (
            [slug] =&gt; grammar!semantic
            [parent] =&gt; grammar
            [name] =&gt; semantic
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; 
        )

    [grammar!syntactic] =&gt; Array
        (
            [slug] =&gt; grammar!syntactic
            [parent] =&gt; grammar
            [name] =&gt; syntactic
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; 
        )

    [grammars] =&gt; Array
        (
            [slug] =&gt; grammars
            [parent] =&gt; 
            [name] =&gt; grammars
            [canonical] =&gt; grammar
            [see] =&gt; 
            [description] =&gt; 
        )

    [graph identifier] =&gt; Array
        (
            [slug] =&gt; graph identifier
            [parent] =&gt; 
            [name] =&gt; graph identifier
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; 
        )

    [graph search] =&gt; Array
        (
            [slug] =&gt; graph search
            [parent] =&gt; 
            [name] =&gt; graph search
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; 
        )

    [graphical information systems] =&gt; Array
        (
            [slug] =&gt; graphical information systems
            [parent] =&gt; 
            [name] =&gt; graphical information systems
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; 
        )

    [graphics processing unit] =&gt; Array
        (
            [slug] =&gt; graphics processing unit
            [parent] =&gt; 
            [name] =&gt; graphics processing unit
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; 
        )

    [grey matter] =&gt; Array
        (
            [slug] =&gt; grey matter
            [parent] =&gt; 
            [name] =&gt; grey matter
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; Grey matter is the material in the outer parts of of the brain as compared with the {{white matter}} towards the center.  The majority of the cell bodies of neorones is within the grey matter and connections within the grey matter are largely local.  The nerve fibres that connect more distant neorones, extend through the {{white matter}}.
        )

    [grey scale] =&gt; Array
        (
            [slug] =&gt; grey scale
            [parent] =&gt; 
            [name] =&gt; grey scale
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; 
        )

    [grey scales] =&gt; Array
        (
            [slug] =&gt; grey scales
            [parent] =&gt; 
            [name] =&gt; grey scales
            [canonical] =&gt; grey scale
            [see] =&gt; 
            [description] =&gt; 
        )

    [grey-scale] =&gt; Array
        (
            [slug] =&gt; grey-scale
            [parent] =&gt; 
            [name] =&gt; grey-scale
            [canonical] =&gt; grey scale
            [see] =&gt; 
            [description] =&gt; 
        )

    [grey-scale image] =&gt; Array
        (
            [slug] =&gt; grey-scale image
            [parent] =&gt; 
            [name] =&gt; grey-scale image
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; 
        )

    [grey-scale images] =&gt; Array
        (
            [slug] =&gt; grey-scale images
            [parent] =&gt; 
            [name] =&gt; grey-scale images
            [canonical] =&gt; grey-scale image
            [see] =&gt; 
            [description] =&gt; 
        )

    [grices's conversational maxims] =&gt; Array
        (
            [slug] =&gt; grices's conversational maxims
            [parent] =&gt; 
            [name] =&gt; Grices's conversational maxims
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; As part of his pragmatic theory of language, Paul Grice proposed four maxims that were implicitly adopted during most day-to-day human cooperative interactions: &lt;br&gt;&lt;em&gt;quantity&lt;/em&gt; &amp;ndash; try to give as much information as is necessary, but no more; &lt;br&gt;&lt;em&gt;quality&lt;/em&gt; &amp;ndash; try to offer truthful and reliable information; &lt;br&gt;&lt;em&gt;relation&lt;/em&gt; &amp;ndash; try to offer relevant and pertinent information; &lt;br&gt;&lt;em&gt;manner&lt;/em&gt; &amp;ndash; try to be as clear and broef as possible whilst avoiding ambiguity. &lt;br&gt; The overacrching pronciple is that people try to offer infornation that helps to further the conversation.
        )

    [ground truth] =&gt; Array
        (
            [slug] =&gt; ground truth
            [parent] =&gt; 
            [name] =&gt; ground truth
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; 
        )

    [hadoop] =&gt; Array
        (
            [slug] =&gt; hadoop
            [parent] =&gt; 
            [name] =&gt; Hadoop
            [canonical] =&gt; 
            [see] =&gt; apache hadoop
            [description] =&gt; 
        )

    [hal] =&gt; Array
        (
            [slug] =&gt; hal
            [parent] =&gt; 
            [name] =&gt; hal
            [canonical] =&gt; hal 9000
            [see] =&gt; 
            [description] =&gt; 
        )

    [hal 9000] =&gt; Array
        (
            [slug] =&gt; hal 9000
            [parent] =&gt; 
            [name] =&gt; HAL 9000
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; HAL 9000 was the computer in the film &quot;2001 A Space Odyssey&quot;, which developed {{emotions}} and {{consciousness}}.  In the film it becomes afraid when it makes a mistake, and then murderously paranoid fearing.  HAL is evetually turned off, but as this happens pleads for its continuing existence.
        )

    [handwriting recognition] =&gt; Array
        (
            [slug] =&gt; handwriting recognition
            [parent] =&gt; 
            [name] =&gt; handwriting recognition
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; 
        )

    [hard problem of consciousness] =&gt; Array
        (
            [slug] =&gt; hard problem of consciousness
            [parent] =&gt; 
            [name] =&gt; hard problem of consciousness
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; The hard problem of consciousness is a term coined by philosopher David Chalmers.  It refers to the bare experience of bieng as opposed to the 'easy' problems of consciousness such as keeping track of one's mental states.
        )

    [hard threshold] =&gt; Array
        (
            [slug] =&gt; hard threshold
            [parent] =&gt; 
            [name] =&gt; hard threshold
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; 
        )

    [hardy, thomas] =&gt; Array
        (
            [slug] =&gt; hardy, thomas
            [parent] =&gt; 
            [name] =&gt; Hardy, Thomas
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; Thomas Hardy (1840-1928) was a novelists famous for his bleak plotlines.
        )

    [Herbert A. Simon] =&gt; Array
        (
            [slug] =&gt; Herbert A. Simon
            [parent] =&gt; 
            [name] =&gt; Herbert A. Simon
            [canonical] =&gt; 
            [see] =&gt; Simon, Herbert
            [description] =&gt; 
        )

    [heteroassociative memory] =&gt; Array
        (
            [slug] =&gt; heteroassociative memory
            [parent] =&gt; 
            [name] =&gt; heteroassociative memory
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; 
        )

    [heterogeneous events] =&gt; Array
        (
            [slug] =&gt; heterogeneous events
            [parent] =&gt; 
            [name] =&gt; heterogeneous events
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; 
        )

    [heterogeneous sources] =&gt; Array
        (
            [slug] =&gt; heterogeneous sources
            [parent] =&gt; 
            [name] =&gt; heterogeneous sources
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; 
        )

    [heuristic evaluation] =&gt; Array
        (
            [slug] =&gt; heuristic evaluation
            [parent] =&gt; 
            [name] =&gt; heuristic evaluation
            [canonical] =&gt; heuristic evaluation function
            [see] =&gt; 
            [description] =&gt; 
        )

    [heuristic evaluation function] =&gt; Array
        (
            [slug] =&gt; heuristic evaluation function
            [parent] =&gt; 
            [name] =&gt; heuristic evaluation function
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; An heuristic evaluation function is some form of guess at an intermediate state about the eventual outcome of a search.  This is  used in a {{heuristic search}} to act as a guide as to the best order to explore many different options.  For example, whilst searching for shortest routes between places, we might use the {{Euclidean distance}} between two cities as a heuristic for the eventual shortest road distance well before we explore the potential routes between them.
        )

    [heuristic search] =&gt; Array
        (
            [slug] =&gt; heuristic search
            [parent] =&gt; 
            [name] =&gt; heuristic search
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; 
        )

    [heuristics] =&gt; Array
        (
            [slug] =&gt; heuristics
            [parent] =&gt; 
            [name] =&gt; heuristics
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; 
        )

    [hick's law] =&gt; Array
        (
            [slug] =&gt; hick's law
            [parent] =&gt; 
            [name] =&gt; Hick's law
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; Hick's Law is one of the few results in cognitve psychology with a precise formulation, namely that the time taken to mae a choice increases with the logarithmof the number of choices.  That is, if someone has to make a choice between N alternatves, the reaction time  T is given by:&lt;br&gt; &amp;nbsp; &amp;nbsp; &amp;nbsp; T &amp;nbsp; &amp;prop; &amp;nbsp; log( N ) &lt;br&gt;&lt;br&gt; In fact, the original experiments that gave to this were based on highly structured inputs that enabled 'binarychop' style decison strategies.  Given a long list of randomly sorted alternatives there is at least a linear time to read the list.  However, whether the exact time is logarithmic or linear, the qulaitiave result still holds: the more alterbatuves the longer ot takes a person to kae the decsions. &lt;br&gt;&lt;br&gt; This isimportant when designing intelligent {{human-in-the-loop}} systems, as it is sensible to present muliple alternatives to the human, but not too many.
        )

    [hidden markov model] =&gt; Array
        (
            [slug] =&gt; hidden markov model
            [parent] =&gt; 
            [name] =&gt; hidden Markov model
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; A hidden Markov model (HMM) extends the {{Markov model}} by adding a hidden state.  This can be used to model natural phenomena, such as weather systems that ave some sort of longer term meteorological state as well as the immediately measurable rain or shine.   Formally, given a {[time series}} or sequence of observable events/tokens, while a simple Markov model has a matrix of {{probabilities}} M&lt;sub&gt;jk&lt;/sub&gt; giving the probability that event j comes after event k; instead the hidde Markov model has more complex matrix H&lt;sub&gt;js;kt&lt;/sub&gt; which gives the probabilty that given an observable event j and (unobservable) hidden state s, the next event will be k and the hidden state t.  &lt;br&gt;&lt;br&gt;. Infering a HMM from traing sequences is more diffocult than training a plain Markov model as the hidden state isn not avaiable in the trainng data and so must itself be {{ftted}} as part of the training process. &lt;br&gt;&lt;br.&gt; Note that {{higher order Markiv modles}} (those that take into account several past observable events), which are easier to learns, can always be represented as HMMs by simply taking the hidden state to be the past tokens.  This can be one way the {{bootstrap}} the HMM training process.
        )

    [hidden markov models] =&gt; Array
        (
            [slug] =&gt; hidden markov models
            [parent] =&gt; 
            [name] =&gt; hidden markov models
            [canonical] =&gt; hidden markov model
            [see] =&gt; 
            [description] =&gt; 
        )

    [hidden state] =&gt; Array
        (
            [slug] =&gt; hidden state
            [parent] =&gt; 
            [name] =&gt; hidden state
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; 
        )

    [hierarchical grammars] =&gt; Array
        (
            [slug] =&gt; hierarchical grammars
            [parent] =&gt; 
            [name] =&gt; hierarchical grammars
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; 
        )

    [high-level programming languages] =&gt; Array
        (
            [slug] =&gt; high-level programming languages
            [parent] =&gt; 
            [name] =&gt; high-level programming languages
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; The term high-level programming language is deisgned to be closer the the human problem solution than an assembly language or similar that are operate very close to the computers raw way of dealing with values and code.  High levels languages usually include constructs for loops, functions and objects as well as ways to manipulate individual values and arrays; common examples include {{Python}} and {{Java}}.  However, there are other kinds of high-level programming language including {{logic programong}} such as {{Prolog}} or {{fucntional languages}}, such as {{Lisp}}. 
        )

    [higher-order function] =&gt; Array
        (
            [slug] =&gt; higher-order function
            [parent] =&gt; 
            [name] =&gt; higher-order function
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; A higher-order function is a function where one or more if its parameters is itself a function.  The {{map}} and {{reduce}} funcions that were the inspration for {{MapReduce}} are both examples of higher order functions.
        )

    [hill climbing] =&gt; Array
        (
            [slug] =&gt; hill climbing
            [parent] =&gt; 
            [name] =&gt; hill climbing
            [canonical] =&gt; hill climbing algorithm
            [see] =&gt; 
            [description] =&gt; 
            [subentries] =&gt; Array
                (
                    [hill climbing!forgetful] =&gt; Array
                        (
                            [slug] =&gt; hill climbing!forgetful
                            [parent] =&gt; hill climbing
                            [name] =&gt; forgetful
                            [canonical] =&gt; 
                            [see] =&gt; forgetful hill climbing
                            [description] =&gt; 
                        )

                )

        )

    [hill climbing algorithm] =&gt; Array
        (
            [slug] =&gt; hill climbing algorithm
            [parent] =&gt; 
            [name] =&gt; hill climbing algorithm
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; Hill climbing is one of the simplest {{optimisation}} algorithms.  On a {{search graph}} one starts at a state, and then looks at neighbouring states choosing a better one and then shifting focus to that state.  This step-by-step process is repeated until there are no neighbouring states that are better, at which point one has reached a {{local maximum}}, but not necessarily the {{global maximum}}.   The simplest form of hill climb always chooses the very best neighbouring state, but if there are many neighbouring states, or when the search is over a continuous search space, it may be necessary to choose any good-enough next step that improves on the current state.  &lt;br&gt;&lt;br&gt; Note the name 'hill climbing' suggests that the optimal solutuon is the one with highest score, but there are also many applications where the smallest score is requred, for example, distance to a target state.  In these case it is more like 'valley seeking', but the term 'hill climbing' is still applied.  Note too that in some searches, particularly {{tree search}} some states are not goal states in themselves and so do not have a score or fitness as such, but instead have a {{heuristic}} value.  In these cases one always moves to the best (or good enough) neighbour, even if this is actually worse than the heuristic of the current state.
        )

    [hill climbing!forgetful] =&gt; Array
        (
            [slug] =&gt; hill climbing!forgetful
            [parent] =&gt; hill climbing
            [name] =&gt; forgetful
            [canonical] =&gt; 
            [see] =&gt; forgetful hill climbing
            [description] =&gt; 
        )

    [hill climbing with backtracking] =&gt; Array
        (
            [slug] =&gt; hill climbing with backtracking
            [parent] =&gt; 
            [name] =&gt; hill climbing with backtracking
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; Simple {{forgetful hill climbing}} constantly tries to look foer neighbouring places or states that are better than the surrent state, but keeps no track of where it has been.   By adding {{backtracking}} hill climbing can deal impasses, when it cannot find suitable next states, or with {{local maxima}} by restarting from a previous state.  If the hill-climbing is {{deterministic}} (chooisng the very best next step) the algorithm also has to keep track of what choices it made in the past otherwise it would simply repeat the same steps after backtracking!  This is less important for {{stochasitic}} variants of hillclimbing (such as {{simulated annealing}}).
        )

    [hill-climbing] =&gt; Array
        (
            [slug] =&gt; hill-climbing
            [parent] =&gt; 
            [name] =&gt; hill-climbing
            [canonical] =&gt; hill climbing algorithm
            [see] =&gt; 
            [description] =&gt; 
        )

    [historical representation] =&gt; Array
        (
            [slug] =&gt; historical representation
            [parent] =&gt; 
            [name] =&gt; historical representation
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; In {{robotics}} a historical representation keeps track of the hstory of observationsn and actions.  Unlike a {{current state representation}}, it is able to re-assess past observations given new evidence, but it requres more memor and computation to periodically re-interpret the history.
        )

    [hmm] =&gt; Array
        (
            [slug] =&gt; hmm
            [parent] =&gt; 
            [name] =&gt; HMM
            [canonical] =&gt; 
            [see] =&gt; hidden markov model
            [description] =&gt; 
        )

    [hold out] =&gt; Array
        (
            [slug] =&gt; hold out
            [parent] =&gt; 
            [name] =&gt; hold out
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; 
        )

    [homeostasis] =&gt; Array
        (
            [slug] =&gt; homeostasis
            [parent] =&gt; 
            [name] =&gt; homeostasis
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; Homeostasis is the tendency of living things to mantain a healthy or desired  state.  This applies to internal state such as body temperarure --  humans shiver to generate heat through muscle movement when too cold; and dogs pant to lose heat through evaporation when too hot.  It also applies to external state such as the small balancing movements we make when standing.  Similarly {{robots}} and forms of {{artificial life}} need to maintain 'healthy' state, such as not overheating or running out of energy.
        )

    [homogeneous events] =&gt; Array
        (
            [slug] =&gt; homogeneous events
            [parent] =&gt; 
            [name] =&gt; homogeneous events
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; 
        )

    [hopfield networks] =&gt; Array
        (
            [slug] =&gt; hopfield networks
            [parent] =&gt; 
            [name] =&gt; Hopfield networks
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; A Hopfield network is a totally connected {{neural network}} often used as {{autoencoder}}.  Like {{Boltzmann machines}}, the Hopfield network is related to {[spin glass models}} in physics.  Hopfield networks differ from Boltzmann machines principally in the update model for node weights, notably, while Hopfield networks adopt a continuous learning mechanism with smooth {{thresholds}}, similar to that in {{backpropogation}}; {{Boltzmann machines}} use {{stochastic}} firing that effectively creates a {{Markov process}}, and are more similar to {{spiking neural networks}}.
        )

    [human computation] =&gt; Array
        (
            [slug] =&gt; human computation
            [parent] =&gt; 
            [name] =&gt; human computation
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; 
        )

    [human in the loop] =&gt; Array
        (
            [slug] =&gt; human in the loop
            [parent] =&gt; 
            [name] =&gt; human in the loop
            [canonical] =&gt; 
            [see] =&gt; human-in-the-loop
            [description] =&gt; 
        )

    [human brain project] =&gt; Array
        (
            [slug] =&gt; human brain project
            [parent] =&gt; 
            [name] =&gt; human brain project
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; 
        )

    [human-computer interaction] =&gt; Array
        (
            [slug] =&gt; human-computer interaction
            [parent] =&gt; 
            [name] =&gt; Human&amp;nbsp;Computer Interaction
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; Human--Computer Interaction (HCI) is the study of the way people interact with technology.  It is both an academic area and also a design discipline that includes {{user experince design}} and {{user interace design}}.  Within AI it may be an application area, in particular {{intelligent user interfaces}}.  More generally whenevr an AI system is used by people there needs to be effective design adopting both general {{HCI}} principles, but also special design issues for human--AI interactions.
        )

    [human-in-the-loop] =&gt; Array
        (
            [slug] =&gt; human-in-the-loop
            [parent] =&gt; 
            [name] =&gt; human-in-the-loop
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; A human-in-the-loop system si one where a human makes or strongly influences the final decsion.  For example, in some {{safety-ciritical}} situations, such as medicine or air traffic control, there may be professional or regulatory reasons way a human must oversee a final decision.  It is particularly important in {{autonomous weapons}}, where many argue that the final decison to fire shoudl come from a human.  Sometimes the term is applied more genrallly when the AI includes human participants, for exmaple online AI-driven chat whcuh may pass more complex queries to  human, altough arguably this is more properly egarded as {{human computation}}.
        )

    [hybrid] =&gt; Array
        (
            [slug] =&gt; hybrid
            [parent] =&gt; 
            [name] =&gt; hybrid
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; Hybrid or hybrid system is about things that involve different knds of components, and is encountered in several ways within AI..  {{Hybrid AI}} may refer to {{hybrid architectures}} employing different kinds of AI in a single system, or {{hybrid human-AI}} applications that employ AI closely alongside humans as a form of partnership.   It may also refer to {{hybrid problems}} that include both discrete and continuous parameters.
        )

    [hybrid ai] =&gt; Array
        (
            [slug] =&gt; hybrid ai
            [parent] =&gt; 
            [name] =&gt; hybrid AI
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; {{Hybrid AI}} may refer to {{hybrid architectures}} employing different kinds of AI in a single system, or {{hybrid human-AI}} applications that employ AI closely alongside humans as a form of partnership.  
        )

    [hybrid architecture] =&gt; Array
        (
            [slug] =&gt; hybrid architecture
            [parent] =&gt; 
            [name] =&gt; hybrid architecture
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; An hybrid architecture is one where different moduels or layers use different kinds of AI technolgies.  For example, we might use a {{random forest}} to perform low-level classification (say of human activity in a care setting), and then use these classes in a hand-crafted {{rule-based system}} (say &quot;if suspected fall, alert warden&quot;).
        )

    [hybrid architectures] =&gt; Array
        (
            [slug] =&gt; hybrid architectures
            [parent] =&gt; 
            [name] =&gt; hybrid architectures
            [canonical] =&gt; hybrid architecture
            [see] =&gt; 
            [description] =&gt; 
        )

    [hybrid expert system] =&gt; Array
        (
            [slug] =&gt; hybrid expert system
            [parent] =&gt; 
            [name] =&gt; hybrid expert system
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; A hybrid expert system is an {{expert system}} in which the AI system and expert continue to operate alongside each other after the {{training phase}}.  For example, a medical expert system might offer advice or suggest diagnoses, which the doctor might use to inform theor decison, or possibly respond by inputing additional infornation about the pattient, which may then change the expert system's advice.  Hybrid expert systems are a form of {{hybrid human-AI} system.
        )

    [hybrid expert systems] =&gt; Array
        (
            [slug] =&gt; hybrid expert systems
            [parent] =&gt; 
            [name] =&gt; hybrid expert systems
            [canonical] =&gt; hybrid expert systems
            [see] =&gt; 
            [description] =&gt; 
        )

    [hybrid human-AI] =&gt; Array
        (
            [slug] =&gt; hybrid human-AI
            [parent] =&gt; 
            [name] =&gt; hybrid human&amp;ndash;AI
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; Hybrid human--AI systems are those in whch humans and AI work closely alongside each other.  While all AI systems interact with human systems at some pount, this may be quite indirect or distant, for example, an engine monitoring system in a car or and AI system creating a new medicine.  In hybrid human--A systems this connection is more direct, for example, handing control back and forth in a semi-autonomous car, or a judge using the (controversial) {{COMPAS}} system's advice in a parole hearing.  Even within hybrid human--AI systems there can be levels of cooperation, from situations where one or other is clearly the dominant, to those where it is more like a two-way cooperation, as in some forms of {{decision-support system}}.
        )

    [hybrid problems] =&gt; Array
        (
            [slug] =&gt; hybrid problems
            [parent] =&gt; 
            [name] =&gt; hybrid problems
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; Hybrid problems are those that involve both {{continuous}} and {{discrete}} parameters or processes.  At is simplest this may be a dataset some {{continuous valued}} features and {{discrete valued}} features.  It may also refer to underlying systems that include both physical dynamics and digital control.  In the {{formal methods}} community the term 'hybrid systems' refers to the latter.  On the whole, designing algorithms to deal with &lt;em&gt;both&lt;/em&gt; continuous and discrete aspects is harder than either on their own.
        )

    [hybrid systems] =&gt; Array
        (
            [slug] =&gt; hybrid systems
            [parent] =&gt; 
            [name] =&gt; hybrid systems
            [canonical] =&gt; 
            [see] =&gt; hybrid ai
            [description] =&gt; 
        )

    [hypertext] =&gt; Array
        (
            [slug] =&gt; hypertext
            [parent] =&gt; 
            [name] =&gt; hypertext
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; Hypertext is when portions of text wihin one or more documents link to one another.  The source location is often called an 'anchor', hence the tag &quot;&amp;lt;a&amp;gt;&quot; in {{HTML}}.  The target of this can point to whole document links, as commonly found in the {{web}} or this glossary, but can also include richer connections.    For example, several early hypertext systems offered multiple outward links from a single anchor point, say, to offer different links for people with different kinds expertise (engineers or users), or different levels of knowledge (high-school or university students).  Often hypertext systems, such as the {{web}} include different media such as images or videos, and the term {{hypermedia}} may be used to distinguish those where this is a major aspect of the information.  &lt;br&gt;&lt;br&gt; Within AI, hypertexts may be a source of knowledge (as in the way Wikipedia has been mined to create the {{dbPedia}} dataset), as an application to support in an {{intelligent user interface}} (for example, tracking common paths through the system and suggestingthese to future users), or as an output when links or indeed parts of the text are autimatically generated.
        )

    [hypothesis-driven reasoning] =&gt; Array
        (
            [slug] =&gt; hypothesis-driven reasoning
            [parent] =&gt; 
            [name] =&gt; hypothesis-driven reasoning
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; 
        )

    [ibm] =&gt; Array
        (
            [slug] =&gt; ibm
            [parent] =&gt; 
            [name] =&gt; IBM
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; 
        )

    [ibm watson] =&gt; Array
        (
            [slug] =&gt; ibm watson
            [parent] =&gt; 
            [name] =&gt; IBM Watson
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; Watson is a computer system designed by {{IBM}} to tackle general knowledge style queries.  It combines  {{natural language processing}}, {{machine learning}} and rapid access to large quantities of text from  encyclopaedias, news feeds and other human-readable data sources.   Watson is famous for defeating human contestents in the US general knowldge TV quiz show {{Jeopardy!}} in 2010.  Since then it has been adapted for other domains including medical information. 
        )

    [ibm's watson] =&gt; Array
        (
            [slug] =&gt; ibm's watson
            [parent] =&gt; 
            [name] =&gt; ibm's watson
            [canonical] =&gt; ibm watson
            [see] =&gt; 
            [description] =&gt; 
        )

    [id3] =&gt; Array
        (
            [slug] =&gt; id3
            [parent] =&gt; 
            [name] =&gt; ID3
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; ID3 (Iterative Dichotomiser 3) and its later variants including {{C4.5}} of concept learning, developed by {{Ross Quinlan}}.  ID3 builds a {{decision tree}} in a top-down fashion using {{information measures}}, typically {{entropy}}, to choose a top level decision and the inductively use the same princple for lower branches.  The simplest version of ID3 simply measures the entropy gain of the decision by counting the frequency examples of each class in the branches.  As the name suggests, basic ID3 was developed for {{binary decisions}}, but there are modifications for {{non-binary decsions}}). There are also modifications to make the trees comprehensible  (see {{expaliable AI}}), either by modifying the inductive learning algorithm itself or {{pruning}} the tree once constructed.
        )

    [identity theft] =&gt; Array
        (
            [slug] =&gt; identity theft
            [parent] =&gt; 
            [name] =&gt; identity theft
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; 
        )

    [if this then that] =&gt; Array
        (
            [slug] =&gt; if this then that
            [parent] =&gt; 
            [name] =&gt; if this then that
            [canonical] =&gt; 
            [see] =&gt; ifttt
            [description] =&gt; 
        )

    [ifttt] =&gt; Array
        (
            [slug] =&gt; ifttt
            [parent] =&gt; 
            [name] =&gt; IFTTT 
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; IFTTT (if this then that) is a commercial web-based systems that allows end-users to configure rules to fire when certain events happen and conditions are met.  For example, you could configure a rule to send you a message when a particular social media chennel mentions a key term.
        )

    [iir] =&gt; Array
        (
            [slug] =&gt; iir
            [parent] =&gt; 
            [name] =&gt; IIR
            [canonical] =&gt; infinite impulse response
            [see] =&gt; 
            [description] =&gt; 
        )

    [image recognition] =&gt; Array
        (
            [slug] =&gt; image recognition
            [parent] =&gt; 
            [name] =&gt; image recognition
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; 
        )

    [image thresholding] =&gt; Array
        (
            [slug] =&gt; image thresholding
            [parent] =&gt; 
            [name] =&gt; image thresholding
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; 
        )

    [image understanding] =&gt; Array
        (
            [slug] =&gt; image understanding
            [parent] =&gt; 
            [name] =&gt; image understanding
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; 
        )

    [imagination] =&gt; Array
        (
            [slug] =&gt; imagination
            [parent] =&gt; 
            [name] =&gt; imagination
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; 
        )

    [imitation game] =&gt; Array
        (
            [slug] =&gt; imitation game
            [parent] =&gt; 
            [name] =&gt; imitation game
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; 
        )

    [implicit feedback] =&gt; Array
        (
            [slug] =&gt; implicit feedback
            [parent] =&gt; 
            [name] =&gt; implicit feedback
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; 
        )

    [incidental] =&gt; Array
        (
            [slug] =&gt; incidental
            [parent] =&gt; 
            [name] =&gt; incidental
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; 
        )

    [independent variable] =&gt; Array
        (
            [slug] =&gt; independent variable
            [parent] =&gt; 
            [name] =&gt; independent variable
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; In {{linear regression}} and other foms of modelling an {{independent variable}} is a value that you can control or is an input to the model.  In contrast to a {{dependent variable}}, which is one of the  model's outputs.  That is the model is effectively of the form: &lt;br&gt; &amp;nbsp; &amp;nbsp; &amp;nbsp; dependent variables &amp;nbsp; = &amp;nbsp;  model( parameters, independent variables ) &lt;br&gt; where one then tries to fit the parameters.
        )

    [induction] =&gt; Array
        (
            [slug] =&gt; induction
            [parent] =&gt; 
            [name] =&gt; induction
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; 
        )

    [inductive learning] =&gt; Array
        (
            [slug] =&gt; inductive learning
            [parent] =&gt; 
            [name] =&gt; inductive learning
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; 
        )

    [inductive reasoning] =&gt; Array
        (
            [slug] =&gt; inductive reasoning
            [parent] =&gt; 
            [name] =&gt; inductive reasoning
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; 
        )

    [industry 4.0] =&gt; Array
        (
            [slug] =&gt; industry 4.0
            [parent] =&gt; 
            [name] =&gt; Industry 4.0
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; Industry 4.0, sometimes also called 'fourth Industrial revolution', refers to the way in which factories and other industrial processes increasingly use smart machines ands sensor-fileld environments to monitor and control processes.
        )

    [inexact search] =&gt; Array
        (
            [slug] =&gt; inexact search
            [parent] =&gt; 
            [name] =&gt; inexact search
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; 
        )

    [infinite impulse response] =&gt; Array
        (
            [slug] =&gt; infinite impulse response
            [parent] =&gt; 
            [name] =&gt; infinite impulse response
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; In an infinite input response (IIR) systemthe current response may be influenced by observations or events arbitrarily far in the past. Typically this arises when ther is some form of feedback loop of past outputs of the system as part of the current input or where there is internal state.  Some statistical time series analysis techniques are IIR such as  a {{auto-regressive}} filters as well as neural network techniqus such as {{recurrent neural networks}}.
        )

    [information] =&gt; Array
        (
            [slug] =&gt; information
            [parent] =&gt; 
            [name] =&gt; information
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; 
        )

    [information entropy] =&gt; Array
        (
            [slug] =&gt; information entropy
            [parent] =&gt; 
            [name] =&gt; information entropy
            [canonical] =&gt; 
            [see] =&gt; entropy
            [description] =&gt; 
        )

    [information gain] =&gt; Array
        (
            [slug] =&gt; information gain
            [parent] =&gt; 
            [name] =&gt; information gain
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; You may be familiar with the &lt;em&gt;base&lt;/em&gt; 2  {{logarithm}} of a number N (written log&lt;sub&gt;2&lt;/sub&gt;(N), or log2(N).  This is defined to be the number L such that N = 2&lt;sup&gt;L&lt;/sup&gt;.  For example,  log&lt;sub&gt;2&lt;/sub&gt;(32) = 5 and log&lt;sub&gt;2&lt;/sub&gt;(10) &amp;tilde; 3.322.  You may also have come across logarithms base 10, so that log&lt;sub&gt;10&lt;/sub&gt;(1000) = 3.   The &lt;em&gt;base of a logarithm&lt;/em&gt; is the generalisation of this; so that if B is the base, then the logarithm v=base B of a nuymber N (written  log&lt;sub&gt;B&lt;/sub&gt;(N)) satisfies: &lt;br&gt;  &amp;nbsp;  &amp;nbsp;  &amp;nbsp;  N  &amp;nbsp; =  &amp;nbsp;  B&lt;sup&gt;log&lt;sub&gt;B&lt;/sub&gt;(N)&lt;/sup&gt; &lt;br&gt; &lt;/br&gt;As well as logarithms base 2 (common in comouting becasue of the use of the binary system) and base 10, you are lielly to come across what are known as the {{natural logarithm}} that is the logarithm to base e, where e is {{Eular's number}} &amp;tilde; 2.7182.  This doesn't sound like a very 'natural' base fo a logarithm, but they occur commonly on physical and bilogical systems. &lt;br&gt;&lt;br&gt; The choice of base only affects the scale of the locgarithm as all logarithms are linearly related to one another and in general given two logarithm bases A and B:&lt;br&gt;  &amp;nbsp;  &amp;nbsp;  &amp;nbsp;  log&lt;sub&gt;A&lt;/sub&gt;(N) &amp;nbsp; = &amp;nbsp;   log&lt;sub&gt;A&lt;/sub&gt;(B) &amp;times;  log&lt;sub&gt;B&lt;/sub&gt;(N)
        )

    [information gain ratio] =&gt; Array
        (
            [slug] =&gt; information gain ratio
            [parent] =&gt; 
            [name] =&gt; information gain ratio
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; Information gain ratio is an {{information measure}} used in {{decision tree}} bulding algorithms, notably {{C4.5}}, that tries to remove some of the bias towards decsions with lareg numbers of alternatives.  It is obtained by dividing the raw {{information gain}} (gain(C,G)) by the entropy of the decision grouping.  That is &lt;br&gt; &amp;nbsp; Information gain ratio =  gain(C,G) / H(G) &lt;br&gt; where H(G) is the entropy of the division into the groups. &lt;br&gt;&lt;br&gt;. This is similar to but different from {{normalised mutual information}}, as the latter divides by H(C)+H(G).   In practice this means the information gain ratio it is weighted slightly more against larger numbers of groups compared with  {{normalised mutual information}}.
        )

    [information preserving] =&gt; Array
        (
            [slug] =&gt; information preserving
            [parent] =&gt; 
            [name] =&gt; information preserving
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; 
        )

    [information processing] =&gt; Array
        (
            [slug] =&gt; information processing
            [parent] =&gt; 
            [name] =&gt; information processing
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; 
        )

    [inhibition] =&gt; Array
        (
            [slug] =&gt; inhibition
            [parent] =&gt; 
            [name] =&gt; inhibition
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; 
        )

    [insight] =&gt; Array
        (
            [slug] =&gt; insight
            [parent] =&gt; 
            [name] =&gt; insight
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; 
        )

    [instabilities] =&gt; Array
        (
            [slug] =&gt; instabilities
            [parent] =&gt; 
            [name] =&gt; instabilities
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; 
        )

    [intellectual property] =&gt; Array
        (
            [slug] =&gt; intellectual property
            [parent] =&gt; 
            [name] =&gt; intellectual property
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; 
        )

    [intelligence] =&gt; Array
        (
            [slug] =&gt; intelligence
            [parent] =&gt; 
            [name] =&gt; intelligence
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; 
            [subentries] =&gt; Array
                (
                    [intelligence!computation view] =&gt; Array
                        (
                            [slug] =&gt; intelligence!computation view
                            [parent] =&gt; intelligence
                            [name] =&gt; computation view
                            [canonical] =&gt; 
                            [see] =&gt; 
                            [description] =&gt; 
                        )

                    [intelligence!emotion] =&gt; Array
                        (
                            [slug] =&gt; intelligence!emotion
                            [parent] =&gt; intelligence
                            [name] =&gt; emotion
                            [canonical] =&gt; 
                            [see] =&gt; 
                            [description] =&gt; 
                        )

                    [intelligence!morals] =&gt; Array
                        (
                            [slug] =&gt; intelligence!morals
                            [parent] =&gt; intelligence
                            [name] =&gt; morals
                            [canonical] =&gt; 
                            [see] =&gt; 
                            [description] =&gt; 
                        )

                    [intelligence!responsibility] =&gt; Array
                        (
                            [slug] =&gt; intelligence!responsibility
                            [parent] =&gt; intelligence
                            [name] =&gt; responsibility
                            [canonical] =&gt; 
                            [see] =&gt; 
                            [description] =&gt; 
                        )

                )

        )

    [intelligence!computation view] =&gt; Array
        (
            [slug] =&gt; intelligence!computation view
            [parent] =&gt; intelligence
            [name] =&gt; computation view
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; 
        )

    [intelligence!emotion] =&gt; Array
        (
            [slug] =&gt; intelligence!emotion
            [parent] =&gt; intelligence
            [name] =&gt; emotion
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; 
        )

    [intelligence!morals] =&gt; Array
        (
            [slug] =&gt; intelligence!morals
            [parent] =&gt; intelligence
            [name] =&gt; morals
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; 
        )

    [intelligence!responsibility] =&gt; Array
        (
            [slug] =&gt; intelligence!responsibility
            [parent] =&gt; intelligence
            [name] =&gt; responsibility
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; 
        )

    [intelligent user interfaces] =&gt; Array
        (
            [slug] =&gt; intelligent user interfaces
            [parent] =&gt; 
            [name] =&gt; intelligent user interfaces
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; 
        )

    [intent] =&gt; Array
        (
            [slug] =&gt; intent
            [parent] =&gt; 
            [name] =&gt; intent
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; 
        )

    [intentional representation] =&gt; Array
        (
            [slug] =&gt; intentional representation
            [parent] =&gt; 
            [name] =&gt; intentional representation
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; For a {{logical expression}} or {{database query}} the intensional representation is the formulae of the expression or the query expresson in a notation such as {{SQL}}.  In constrats the  {{extensional representation}} is the collection of all values that satisfy the expression or are returned by the query.
        )

    [intentionality] =&gt; Array
        (
            [slug] =&gt; intentionality
            [parent] =&gt; 
            [name] =&gt; intentionality
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; In the philosophy of intelligence and consciousness, intentionality refers to the way that words are not merely sounds, but have real menaing.  When a person says &quot;&quot;the sky is blue&quot;&quot;, it is nt just a statement that has been garnered from weather data or observed through the window, but the idea of 'sky' an ''blueness' are attached to a real sense of what the person means by 'blue'
        )

    [interactive voice response] =&gt; Array
        (
            [slug] =&gt; interactive voice response
            [parent] =&gt; 
            [name] =&gt; interactive voice response
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; Interactive voice response (IVR) systems are telephone-based interactive user interfaces which combine voicxe output form the computer with user input in the form of keypresses on the phone or simple human voice responses such as &quot;Yes/No&quot; or speaking digits.  Most systems follow a fixed dialgogue tree, where the user responses determine which path to follow.
        )

    [intercept] =&gt; Array
        (
            [slug] =&gt; intercept
            [parent] =&gt; 
            [name] =&gt; intercept
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; When drawing a straight line and especially in {{linear regression}} o an x--y graph, the intercept is the place where the line crosses the y axis, in other words the value when x=0.. In the case when there is a single {{independent variable}} (say time) and many {{dependent variables}}, the intercept is a vector.
        )

    [internet of things] =&gt; Array
        (
            [slug] =&gt; internet of things
            [parent] =&gt; 
            [name] =&gt; internet of things
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; The internet of things (IoT) refers to the way in which many everyday devices are now internet-enabled and thus part of the global network.  This includes domestic devices (e.g. internet-enabled kettles, smart-fridges, and door-step cameras), {{smart cities}} (e.g. traffic lights, pedestrian flow sensors, pollution monitoring) and industrial applications (e.g. stock monitoring through {{NFC}}, industrial robots).  IoT offers the oppoortunity to interlink these devices and thus allow applications, such as automatically adding milk to your internet shopping list when it runs low in the fridge, or adjusting traffic light timings during epriods of conjestion.  However, there are worries about {{privacy}}  (e.g. if the supermarket learns about your daily eating) and {{security}} (e.g. when building heating systems have been targetted by cybercriminals). 
        )

    [interpret] =&gt; Array
        (
            [slug] =&gt; interpret
            [parent] =&gt; 
            [name] =&gt; interpret
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; 
        )

    [interpreter] =&gt; Array
        (
            [slug] =&gt; interpreter
            [parent] =&gt; 
            [name] =&gt; interpreter
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; 
        )

    [iot] =&gt; Array
        (
            [slug] =&gt; iot
            [parent] =&gt; 
            [name] =&gt; IoT
            [canonical] =&gt; 
            [see] =&gt; internet of things
            [description] =&gt; 
        )

    [isbn] =&gt; Array
        (
            [slug] =&gt; isbn
            [parent] =&gt; 
            [name] =&gt; ISBN
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; An ISBN (International Book Serial Number)is a {{unique identifier}} for a version of a book.   Different editions and paperback and hardback versions of the same book have different ISBNs. 
        )

    [iterative broadening] =&gt; Array
        (
            [slug] =&gt; iterative broadening
            [parent] =&gt; 
            [name] =&gt; iterative broadening
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; 
        )

    [iterative deepening] =&gt; Array
        (
            [slug] =&gt; iterative deepening
            [parent] =&gt; 
            [name] =&gt; iterative deepening
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; 
        )

    [ivr] =&gt; Array
        (
            [slug] =&gt; ivr
            [parent] =&gt; 
            [name] =&gt; IVR
            [canonical] =&gt; 
            [see] =&gt; interactive voice response
            [description] =&gt; 
        )

    [jaccard similarity] =&gt; Array
        (
            [slug] =&gt; jaccard similarity
            [parent] =&gt; 
            [name] =&gt; Jaccard similarity
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; Jaccard similarity is a measure of {{similarity}} between two doucuments.  The Jaccard similarity to compare two documents, doc&lt;sub&gt;1&lt;/sub&gt; and doc&lt;sub&gt;2&lt;/sub&gt;, works on the {{bag of words}} (say words&lt;sub&gt;1&lt;/sub&gt; and words&lt;sub&gt;2&lt;/sub&gt;, and then calculates &lt;br&gt; &amp;nbsp;&amp;nbsp;&amp;nbsp;&amp;nbsp; | words&lt;sub&gt;1&lt;/sub&gt; &amp;cap; words&lt;sub&gt;2&lt;/sub&gt; | / | words&lt;sub&gt;1&lt;/sub&gt; &amp;cup; words&lt;sub&gt;2&lt;/sub&gt; | &lt;br&gt;. That is the number of distinct words in both documents divided by the number of distinct words in the union.  The Jaccard similarity is used heavily in {{document retrieval}} algorithms.
        )

    [jacquard loom] =&gt; Array
        (
            [slug] =&gt; jacquard loom
            [parent] =&gt; 
            [name] =&gt; Jacquard loom
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; The Jacquard loom used punch cards to allow a rich variety of weavng patterns, far beyond fixed methods of tying up that preceded it.  It was the inspiration for the punch cards used in {{Babbage}}'s {{Analytical Engine}} and hence the Jacquard loom can be seen as a direct desc endent of the modern programmable computer. Indeed, these orgins inspired {{Ada Lovelace}} to write about the way the {{Analytical Engine}} &quot;&quot;weaves algebraical patterns just as the Jacquard-loom weaves flowers and leaves&quot;.
        )

    [jacquard-loom] =&gt; Array
        (
            [slug] =&gt; jacquard-loom
            [parent] =&gt; 
            [name] =&gt; jacquard-loom
            [canonical] =&gt; jacquard loom
            [see] =&gt; 
            [description] =&gt; 
        )

    [jeopardy] =&gt; Array
        (
            [slug] =&gt; jeopardy
            [parent] =&gt; 
            [name] =&gt; Jeopardy!
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; Jeopardy! is US general knowldge TV quiz show.  {{IBM Watson}} computer won the game in 2010 against human openents.  It used a combination of natural language processing, machine learning and rapid access to large quantities of text from  encyclopaedias, news feeds and other human-readable data sources.
        )

    [john laird] =&gt; Array
        (
            [slug] =&gt; john laird
            [parent] =&gt; 
            [name] =&gt; John Laird
            [canonical] =&gt; laird, john
            [see] =&gt; 
            [description] =&gt; 
        )

    [john robert anderson] =&gt; Array
        (
            [slug] =&gt; john robert anderson
            [parent] =&gt; 
            [name] =&gt; John Robert Anderson
            [canonical] =&gt; anderson, john robert
            [see] =&gt; 
            [description] =&gt; 
        )

    [john searle] =&gt; Array
        (
            [slug] =&gt; john searle
            [parent] =&gt; 
            [name] =&gt; John Searle
            [canonical] =&gt; searle, john
            [see] =&gt; 
            [description] =&gt; 
        )

    [Joseph Weizenbaum] =&gt; Array
        (
            [slug] =&gt; Joseph Weizenbaum
            [parent] =&gt; 
            [name] =&gt; Joseph Weizenbaum
            [canonical] =&gt; weizenbaum, joseph
            [see] =&gt; 
            [description] =&gt; 
        )

    [junk dna] =&gt; Array
        (
            [slug] =&gt; junk dna
            [parent] =&gt; 
            [name] =&gt; junk DNA
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; The human chromosome contains large tracts of DNA which does not appear to code for anything useful.  This is known as junk DNA.
        )

    [k-fold cross-validation] =&gt; Array
        (
            [slug] =&gt; k-fold cross-validation
            [parent] =&gt; 
            [name] =&gt; k-fold cross-validation
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; K-fold cross-validation is a technique used to validate a {{machine learning}} algorithm for a particular dataset.  The dataset is divided into k pieces (of size N/k, where n is the size of the full dataset) and then the algorithm is trained on k-1 of them with the remaining N/k of the dataset used as test data for an {{accuracy measure}}.  This is then repeated with a different piece left out and the results of the {{accuracy measure}} averaged.
        )

    [k-means] =&gt; Array
        (
            [slug] =&gt; k-means
            [parent] =&gt; 
            [name] =&gt; k-means
            [canonical] =&gt; k-means algorithm
            [see] =&gt; 
            [description] =&gt; 
        )

    [k-means algorithm] =&gt; Array
        (
            [slug] =&gt; k-means algorithm
            [parent] =&gt; 
            [name] =&gt; k-means algorithm
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; K-means is one of the oldest commonly used {{clustering}} algorithms.
        )

    [kasparov, garry] =&gt; Array
        (
            [slug] =&gt; kasparov, garry
            [parent] =&gt; 
            [name] =&gt; Kasparov, Garry
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; Garry Kasparov is a chess grandmaster and world champion for many years, who was famously defeated by the IBM chess programme {{Deep Blue}} in 1996.
        )

    [kernel] =&gt; Array
        (
            [slug] =&gt; kernel
            [parent] =&gt; 
            [name] =&gt; kernel
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; Kernel functions are used in {{machine learning}} including {{support vector machines}} to increas {{non-linear diversity}}.  The term originates in mathematics where a postive definite kernel is a form of {{non-linear}} generalisation of the {{inner prooduct}}.
        )

    [knowledge acquisition] =&gt; Array
        (
            [slug] =&gt; knowledge acquisition
            [parent] =&gt; 
            [name] =&gt; knowledge acquisition
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; 
        )

    [knowledge base] =&gt; Array
        (
            [slug] =&gt; knowledge base
            [parent] =&gt; 
            [name] =&gt; knowledge base
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; 
        )

    [knowledge elicitation] =&gt; Array
        (
            [slug] =&gt; knowledge elicitation
            [parent] =&gt; 
            [name] =&gt; knowledge elicitation
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; 
            [subentries] =&gt; Array
                (
                    [knowledge elicitation!critiquing] =&gt; Array
                        (
                            [slug] =&gt; knowledge elicitation!critiquing
                            [parent] =&gt; knowledge elicitation
                            [name] =&gt; critiquing
                            [canonical] =&gt; 
                            [see] =&gt; 
                            [description] =&gt; 
                        )

                    [knowledge elicitation!focused discussions] =&gt; Array
                        (
                            [slug] =&gt; knowledge elicitation!focused discussions
                            [parent] =&gt; knowledge elicitation
                            [name] =&gt; focused discussions
                            [canonical] =&gt; 
                            [see] =&gt; 
                            [description] =&gt; 
                        )

                    [knowledge elicitation!post-task walkthrough] =&gt; Array
                        (
                            [slug] =&gt; knowledge elicitation!post-task walkthrough
                            [parent] =&gt; knowledge elicitation
                            [name] =&gt; post-task walkthrough
                            [canonical] =&gt; 
                            [see] =&gt; 
                            [description] =&gt; 
                        )

                    [knowledge elicitation!role reversal] =&gt; Array
                        (
                            [slug] =&gt; knowledge elicitation!role reversal
                            [parent] =&gt; knowledge elicitation
                            [name] =&gt; role reversal
                            [canonical] =&gt; 
                            [see] =&gt; 
                            [description] =&gt; 
                        )

                    [knowledge elicitation!seed questions] =&gt; Array
                        (
                            [slug] =&gt; knowledge elicitation!seed questions
                            [parent] =&gt; knowledge elicitation
                            [name] =&gt; seed questions
                            [canonical] =&gt; 
                            [see] =&gt; 
                            [description] =&gt; 
                        )

                    [knowledge elicitation!structured interview] =&gt; Array
                        (
                            [slug] =&gt; knowledge elicitation!structured interview
                            [parent] =&gt; knowledge elicitation
                            [name] =&gt; structured interview
                            [canonical] =&gt; 
                            [see] =&gt; 
                            [description] =&gt; 
                        )

                    [knowledge elicitation!teach-back interview] =&gt; Array
                        (
                            [slug] =&gt; knowledge elicitation!teach-back interview
                            [parent] =&gt; knowledge elicitation
                            [name] =&gt; teach-back interview
                            [canonical] =&gt; 
                            [see] =&gt; 
                            [description] =&gt; 
                        )

                    [knowledge elicitation!think aloud] =&gt; Array
                        (
                            [slug] =&gt; knowledge elicitation!think aloud
                            [parent] =&gt; knowledge elicitation
                            [name] =&gt; think aloud
                            [canonical] =&gt; 
                            [see] =&gt; 
                            [description] =&gt; 
                        )

                    [knowledge elicitation!twenty questions] =&gt; Array
                        (
                            [slug] =&gt; knowledge elicitation!twenty questions
                            [parent] =&gt; knowledge elicitation
                            [name] =&gt; twenty questions
                            [canonical] =&gt; 
                            [see] =&gt; 
                            [description] =&gt; 
                        )

                    [knowledge elicitation!unstructured interview] =&gt; Array
                        (
                            [slug] =&gt; knowledge elicitation!unstructured interview
                            [parent] =&gt; knowledge elicitation
                            [name] =&gt; unstructured interview
                            [canonical] =&gt; 
                            [see] =&gt; 
                            [description] =&gt; 
                        )

                )

        )

    [knowledge elicitation!critiquing] =&gt; Array
        (
            [slug] =&gt; knowledge elicitation!critiquing
            [parent] =&gt; knowledge elicitation
            [name] =&gt; critiquing
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; 
        )

    [knowledge elicitation!focused discussions] =&gt; Array
        (
            [slug] =&gt; knowledge elicitation!focused discussions
            [parent] =&gt; knowledge elicitation
            [name] =&gt; focused discussions
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; 
        )

    [knowledge elicitation!post-task walkthrough] =&gt; Array
        (
            [slug] =&gt; knowledge elicitation!post-task walkthrough
            [parent] =&gt; knowledge elicitation
            [name] =&gt; post-task walkthrough
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; 
        )

    [knowledge elicitation!role reversal] =&gt; Array
        (
            [slug] =&gt; knowledge elicitation!role reversal
            [parent] =&gt; knowledge elicitation
            [name] =&gt; role reversal
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; 
        )

    [knowledge elicitation!seed questions] =&gt; Array
        (
            [slug] =&gt; knowledge elicitation!seed questions
            [parent] =&gt; knowledge elicitation
            [name] =&gt; seed questions
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; 
        )

    [knowledge elicitation!structured interview] =&gt; Array
        (
            [slug] =&gt; knowledge elicitation!structured interview
            [parent] =&gt; knowledge elicitation
            [name] =&gt; structured interview
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; 
        )

    [knowledge elicitation!teach-back interview] =&gt; Array
        (
            [slug] =&gt; knowledge elicitation!teach-back interview
            [parent] =&gt; knowledge elicitation
            [name] =&gt; teach-back interview
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; 
        )

    [knowledge elicitation!think aloud] =&gt; Array
        (
            [slug] =&gt; knowledge elicitation!think aloud
            [parent] =&gt; knowledge elicitation
            [name] =&gt; think aloud
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; 
        )

    [knowledge elicitation!twenty questions] =&gt; Array
        (
            [slug] =&gt; knowledge elicitation!twenty questions
            [parent] =&gt; knowledge elicitation
            [name] =&gt; twenty questions
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; 
        )

    [knowledge elicitation!unstructured interview] =&gt; Array
        (
            [slug] =&gt; knowledge elicitation!unstructured interview
            [parent] =&gt; knowledge elicitation
            [name] =&gt; unstructured interview
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; 
        )

    [knowledge engineer] =&gt; Array
        (
            [slug] =&gt; knowledge engineer
            [parent] =&gt; 
            [name] =&gt; knowledge engineer
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; A knowledeg engoneer is a professional who works with a {{domain expert}} to elicit and represent the expert's knowledge for use in an {{expert system}} or other form of AI.
        )

    [knowledge] =&gt; Array
        (
            [slug] =&gt; knowledge
            [parent] =&gt; 
            [name] =&gt; knowledge
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; 
            [subentries] =&gt; Array
                (
                    [knowledge!declarative] =&gt; Array
                        (
                            [slug] =&gt; knowledge!declarative
                            [parent] =&gt; knowledge
                            [name] =&gt; declarative
                            [canonical] =&gt; 
                            [see] =&gt; declarative knowledge
                            [description] =&gt; 
                        )

                    [knowledge!domain-independent] =&gt; Array
                        (
                            [slug] =&gt; knowledge!domain-independent
                            [parent] =&gt; knowledge
                            [name] =&gt; domain-independent
                            [canonical] =&gt; 
                            [see] =&gt; domain-independent knowledge
                            [description] =&gt; 
                        )

                    [knowledge!domain-specific] =&gt; Array
                        (
                            [slug] =&gt; knowledge!domain-specific
                            [parent] =&gt; knowledge
                            [name] =&gt; domain-specific
                            [canonical] =&gt; 
                            [see] =&gt; domain-specific knowledge
                            [description] =&gt; 
                        )

                    [knowledge!procedural] =&gt; Array
                        (
                            [slug] =&gt; knowledge!procedural
                            [parent] =&gt; knowledge
                            [name] =&gt; procedural
                            [canonical] =&gt; 
                            [see] =&gt; procedural knowledge
                            [description] =&gt; 
                        )

                )

        )

    [knowledge!declarative] =&gt; Array
        (
            [slug] =&gt; knowledge!declarative
            [parent] =&gt; knowledge
            [name] =&gt; declarative
            [canonical] =&gt; 
            [see] =&gt; declarative knowledge
            [description] =&gt; 
        )

    [knowledge!domain-independent] =&gt; Array
        (
            [slug] =&gt; knowledge!domain-independent
            [parent] =&gt; knowledge
            [name] =&gt; domain-independent
            [canonical] =&gt; 
            [see] =&gt; domain-independent knowledge
            [description] =&gt; 
        )

    [knowledge!domain-specific] =&gt; Array
        (
            [slug] =&gt; knowledge!domain-specific
            [parent] =&gt; knowledge
            [name] =&gt; domain-specific
            [canonical] =&gt; 
            [see] =&gt; domain-specific knowledge
            [description] =&gt; 
        )

    [knowledge!procedural] =&gt; Array
        (
            [slug] =&gt; knowledge!procedural
            [parent] =&gt; knowledge
            [name] =&gt; procedural
            [canonical] =&gt; 
            [see] =&gt; procedural knowledge
            [description] =&gt; 
        )

    [knowledge representation] =&gt; Array
        (
            [slug] =&gt; knowledge representation
            [parent] =&gt; 
            [name] =&gt; knowledge representation
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; 
            [subentries] =&gt; Array
                (
                    [knowledge representation!by logic] =&gt; Array
                        (
                            [slug] =&gt; knowledge representation!by logic
                            [parent] =&gt; knowledge representation
                            [name] =&gt; by logic
                            [canonical] =&gt; 
                            [see] =&gt; 
                            [description] =&gt; 
                        )

                    [knowledge representation!by networks] =&gt; Array
                        (
                            [slug] =&gt; knowledge representation!by networks
                            [parent] =&gt; knowledge representation
                            [name] =&gt; by networks
                            [canonical] =&gt; 
                            [see] =&gt; 
                            [description] =&gt; 
                        )

                    [knowledge representation!clarity] =&gt; Array
                        (
                            [slug] =&gt; knowledge representation!clarity
                            [parent] =&gt; knowledge representation
                            [name] =&gt; clarity
                            [canonical] =&gt; 
                            [see] =&gt; 
                            [description] =&gt; 
                        )

                    [knowledge representation!completeness] =&gt; Array
                        (
                            [slug] =&gt; knowledge representation!completeness
                            [parent] =&gt; knowledge representation
                            [name] =&gt; completeness
                            [canonical] =&gt; 
                            [see] =&gt; 
                            [description] =&gt; 
                        )

                    [knowledge representation!effectiveness] =&gt; Array
                        (
                            [slug] =&gt; knowledge representation!effectiveness
                            [parent] =&gt; knowledge representation
                            [name] =&gt; effectiveness
                            [canonical] =&gt; 
                            [see] =&gt; 
                            [description] =&gt; 
                        )

                    [knowledge representation!efficiency] =&gt; Array
                        (
                            [slug] =&gt; knowledge representation!efficiency
                            [parent] =&gt; knowledge representation
                            [name] =&gt; efficiency
                            [canonical] =&gt; 
                            [see] =&gt; 
                            [description] =&gt; 
                        )

                    [knowledge representation!explicitness] =&gt; Array
                        (
                            [slug] =&gt; knowledge representation!explicitness
                            [parent] =&gt; knowledge representation
                            [name] =&gt; explicitness
                            [canonical] =&gt; 
                            [see] =&gt; 
                            [description] =&gt; 
                        )

                    [knowledge representation!expressiveness] =&gt; Array
                        (
                            [slug] =&gt; knowledge representation!expressiveness
                            [parent] =&gt; knowledge representation
                            [name] =&gt; expressiveness
                            [canonical] =&gt; 
                            [see] =&gt; 
                            [description] =&gt; 
                        )

                    [knowledge representation!frame] =&gt; Array
                        (
                            [slug] =&gt; knowledge representation!frame
                            [parent] =&gt; knowledge representation
                            [name] =&gt; frame
                            [canonical] =&gt; 
                            [see] =&gt; 
                            [description] =&gt; 
                        )

                    [knowledge representation!granularity] =&gt; Array
                        (
                            [slug] =&gt; knowledge representation!granularity
                            [parent] =&gt; knowledge representation
                            [name] =&gt; granularity
                            [canonical] =&gt; 
                            [see] =&gt; 
                            [description] =&gt; 
                        )

                    [knowledge representation!metrics] =&gt; Array
                        (
                            [slug] =&gt; knowledge representation!metrics
                            [parent] =&gt; knowledge representation
                            [name] =&gt; metrics
                            [canonical] =&gt; 
                            [see] =&gt; 
                            [description] =&gt; 
                        )

                    [knowledge representation!procedural] =&gt; Array
                        (
                            [slug] =&gt; knowledge representation!procedural
                            [parent] =&gt; knowledge representation
                            [name] =&gt; procedural
                            [canonical] =&gt; 
                            [see] =&gt; procedural knowledge
                            [description] =&gt; 
                        )

                    [knowledge representation!script] =&gt; Array
                        (
                            [slug] =&gt; knowledge representation!script
                            [parent] =&gt; knowledge representation
                            [name] =&gt; script
                            [canonical] =&gt; 
                            [see] =&gt; 
                            [description] =&gt; 
                        )

                    [knowledge representation!structured] =&gt; Array
                        (
                            [slug] =&gt; knowledge representation!structured
                            [parent] =&gt; knowledge representation
                            [name] =&gt; structured
                            [canonical] =&gt; 
                            [see] =&gt; 
                            [description] =&gt; 
                        )

                )

        )

    [knowledge representation!by logic] =&gt; Array
        (
            [slug] =&gt; knowledge representation!by logic
            [parent] =&gt; knowledge representation
            [name] =&gt; by logic
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; 
        )

    [knowledge representation!by networks] =&gt; Array
        (
            [slug] =&gt; knowledge representation!by networks
            [parent] =&gt; knowledge representation
            [name] =&gt; by networks
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; 
        )

    [knowledge representation!clarity] =&gt; Array
        (
            [slug] =&gt; knowledge representation!clarity
            [parent] =&gt; knowledge representation
            [name] =&gt; clarity
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; 
        )

    [knowledge representation!completeness] =&gt; Array
        (
            [slug] =&gt; knowledge representation!completeness
            [parent] =&gt; knowledge representation
            [name] =&gt; completeness
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; 
        )

    [knowledge representation!effectiveness] =&gt; Array
        (
            [slug] =&gt; knowledge representation!effectiveness
            [parent] =&gt; knowledge representation
            [name] =&gt; effectiveness
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; 
        )

    [knowledge representation!efficiency] =&gt; Array
        (
            [slug] =&gt; knowledge representation!efficiency
            [parent] =&gt; knowledge representation
            [name] =&gt; efficiency
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; 
        )

    [knowledge representation!explicitness] =&gt; Array
        (
            [slug] =&gt; knowledge representation!explicitness
            [parent] =&gt; knowledge representation
            [name] =&gt; explicitness
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; 
        )

    [knowledge representation!expressiveness] =&gt; Array
        (
            [slug] =&gt; knowledge representation!expressiveness
            [parent] =&gt; knowledge representation
            [name] =&gt; expressiveness
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; 
        )

    [knowledge representation!frame] =&gt; Array
        (
            [slug] =&gt; knowledge representation!frame
            [parent] =&gt; knowledge representation
            [name] =&gt; frame
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; 
        )

    [knowledge representation!granularity] =&gt; Array
        (
            [slug] =&gt; knowledge representation!granularity
            [parent] =&gt; knowledge representation
            [name] =&gt; granularity
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; 
        )

    [knowledge representation!metrics] =&gt; Array
        (
            [slug] =&gt; knowledge representation!metrics
            [parent] =&gt; knowledge representation
            [name] =&gt; metrics
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; 
        )

    [knowledge representation!procedural] =&gt; Array
        (
            [slug] =&gt; knowledge representation!procedural
            [parent] =&gt; knowledge representation
            [name] =&gt; procedural
            [canonical] =&gt; 
            [see] =&gt; procedural knowledge
            [description] =&gt; 
        )

    [knowledge representation!script] =&gt; Array
        (
            [slug] =&gt; knowledge representation!script
            [parent] =&gt; knowledge representation
            [name] =&gt; script
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; 
        )

    [knowledge representation!structured] =&gt; Array
        (
            [slug] =&gt; knowledge representation!structured
            [parent] =&gt; knowledge representation
            [name] =&gt; structured
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; 
        )

    [knowledge-rich search] =&gt; Array
        (
            [slug] =&gt; knowledge-rich search
            [parent] =&gt; 
            [name] =&gt; knowledge-rich search
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; A knowledge-rich search uses some form of semantic or structured understanding of the problem space in order to improve the accureacy or speed of {{search}}.  For example, if the problem space is a graph of direct road-travel times between cities, then a knowledege-rich serach might use the crow-flies distance between the cities as a {{heuristic}} to guide the search for a shortest path between two locations.  In contrast a low-knowledge or knowledge-poor approach would just take the distances as path costs without taking into account their geographic nature.
        )

    [kohonen nets] =&gt; Array
        (
            [slug] =&gt; kohonen nets
            [parent] =&gt; 
            [name] =&gt; kohonen nets
            [canonical] =&gt; kohonen network
            [see] =&gt; 
            [description] =&gt; 
        )

    [kohonen network] =&gt; Array
        (
            [slug] =&gt; kohonen network
            [parent] =&gt; 
            [name] =&gt; Kohonen networks
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; A Kohonen network is a {{neural network}} algorithm that creates a {{self-organising map}}.  A grid of nodes (usually 2D regular mesh) is initialised with a vector a eaqch node. Then one by one the training data is compared with the nodes and allocated to the node that is closest.  The node itself is swung to be closer to the new example, as are the nodes immediately surrounding it.  Sometimes those a little way off are negatively weighted (pushed away from) the training set item.  Iterating this process leads to a set of node values on the grid such that close cells tend to have some sort of semantic unity in terms of how they classify the data.
        )

    [kohonen networks] =&gt; Array
        (
            [slug] =&gt; kohonen networks
            [parent] =&gt; 
            [name] =&gt; kohonen networks
            [canonical] =&gt; kohonen network
            [see] =&gt; 
            [description] =&gt; 
        )

    [labelling] =&gt; Array
        (
            [slug] =&gt; labelling
            [parent] =&gt; 
            [name] =&gt; labelling
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; Labelling is the process of giving data items some form of label, often a {{classification}}.  In {{supervised machine learning}}, {{human labelling}} is often necessary for training, but sometimes there may be an automatic labelling, for example, if there are  multiple kinds of sensors, or post-hoc knowledge about outcomes.  The execution phase of {{machine learning}} often involves allocating labels to {{unlabelled data}}.
        )

    [laird, j] =&gt; Array
        (
            [slug] =&gt; laird, j
            [parent] =&gt; 
            [name] =&gt; Laird, J
            [canonical] =&gt; laird, john
            [see] =&gt; 
            [description] =&gt; 
        )

    [laird, john] =&gt; Array
        (
            [slug] =&gt; laird, john
            [parent] =&gt; 
            [name] =&gt; Laird, John
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; John Laird is a Perofessor of Computer Science at Michigan State University, is the creator of the {{SOAR}} {{cognitive architecture}}.
        )

    [lamda] =&gt; Array
        (
            [slug] =&gt; lamda
            [parent] =&gt; 
            [name] =&gt; LaMDA 
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; LaMDA is a conversational {{large-language model}} developed by {{Google}}.
        )

    [laplacian operator] =&gt; Array
        (
            [slug] =&gt; laplacian operator
            [parent] =&gt; 
            [name] =&gt; Laplacian operator
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; 
        )

    [laplacian-of-gaussian filter] =&gt; Array
        (
            [slug] =&gt; laplacian-of-gaussian filter
            [parent] =&gt; 
            [name] =&gt; Laplacian-of-Gaussian filter
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; 
        )

    [laplacian-of-gaussian filters] =&gt; Array
        (
            [slug] =&gt; laplacian-of-gaussian filters
            [parent] =&gt; 
            [name] =&gt; laplacian-of-gaussian filters
            [canonical] =&gt; laplacian-of-gaussian filter
            [see] =&gt; 
            [description] =&gt; 
        )

    [large language model meta ai] =&gt; Array
        (
            [slug] =&gt; large language model meta ai
            [parent] =&gt; 
            [name] =&gt; large language model meta AI
            [canonical] =&gt; 
            [see] =&gt; llama
            [description] =&gt; 
        )

    [large language model] =&gt; Array
        (
            [slug] =&gt; large language model
            [parent] =&gt; 
            [name] =&gt; large language model
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; 
        )

    [large language models] =&gt; Array
        (
            [slug] =&gt; large language models
            [parent] =&gt; 
            [name] =&gt; large language models
            [canonical] =&gt; large language model
            [see] =&gt; 
            [description] =&gt; 
        )

    [large-language models] =&gt; Array
        (
            [slug] =&gt; large-language models
            [parent] =&gt; 
            [name] =&gt; large-language models
            [canonical] =&gt; large language model
            [see] =&gt; 
            [description] =&gt; 
        )

    [latent semantic analysis] =&gt; Array
        (
            [slug] =&gt; latent semantic analysis
            [parent] =&gt; 
            [name] =&gt; latent semantic analysis
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; 
        )

    [latent space] =&gt; Array
        (
            [slug] =&gt; latent space
            [parent] =&gt; 
            [name] =&gt; latent space
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; 
        )

    [lateral inhibition] =&gt; Array
        (
            [slug] =&gt; lateral inhibition
            [parent] =&gt; 
            [name] =&gt; lateral inhibition
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; Lateral ihibition is when a groups of nodes at the same layer in a brain or {{artficial neural netwrok}} have interconnections that inhibit one another.  This is may encode aspects where there are different interprettions of a visual scene or data that are effectively fighting for each other.  The blob to the right fo the scene is a pussy cat or a seal, but can't be both at the same time.
        )

    [learning rate] =&gt; Array
        (
            [slug] =&gt; learning rate
            [parent] =&gt; 
            [name] =&gt; learning rate
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; 
        )

    [least-squares] =&gt; Array
        (
            [slug] =&gt; least-squares
            [parent] =&gt; 
            [name] =&gt; least-squares
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; Many {{machine learning}} and {{statistical}} algorithms can be formulated as {{least square}} procedures.  They are effectively minimising the sum of the squares of the differnce between the result they deliver and the desired result -- called the {{mean squared error}}.  In some case this is explicit in the way the algorithm is forumulated, for example the {{backpropagation}} algorithm is a form of {{gradient descent}} that calculates the changes needed to each weight to make the best reduction in the mean square.  Simularly, {{linear rgeression}} minimises the sum of squares of the {{residuals}} (that is the difference between the fitted {[regression line}} and the actual data points).
        )

    [lee sedol] =&gt; Array
        (
            [slug] =&gt; lee sedol
            [parent] =&gt; 
            [name] =&gt; Lee Sedol
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; Lee Se-dol is the Go grandmaster who was defeated by {{AlphaGo}} in 2016.
        )

    [lego-style matching] =&gt; Array
        (
            [slug] =&gt; lego-style matching
            [parent] =&gt; 
            [name] =&gt; lego-style matching
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; 
        )

    [lemoine, blake] =&gt; Array
        (
            [slug] =&gt; lemoine, blake
            [parent] =&gt; 
            [name] =&gt; Lemoine, Blake
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; Blake Lemoine was  a Google engineer who was suspended in 2022 and later fired because he had claimed that the Google chatbot {{LaMDA}} was sentient 
        )

    [lenat, doug] =&gt; Array
        (
            [slug] =&gt; lenat, doug
            [parent] =&gt; 
            [name] =&gt; Lenat, Doug
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; Doug Lenart was a visionary AI researcher who founded the long-running {{CYC project}} to codify and reason about general knowledge.
        )

    [lex] =&gt; Array
        (
            [slug] =&gt; lex
            [parent] =&gt; 
            [name] =&gt; Lex
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; Lex is a notation and Unix command-line programme to generate {{lexical analysers}}.  It uses regular expressions to match valid tokens, such as an variable name or number.  It is usualy used as a pre-processor before syntax analysis.
        )

    [lexical processing] =&gt; Array
        (
            [slug] =&gt; lexical processing
            [parent] =&gt; 
            [name] =&gt; lexical processing
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; 
        )

    [lexicon] =&gt; Array
        (
            [slug] =&gt; lexicon
            [parent] =&gt; 
            [name] =&gt; lexicon
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; 
        )

    [liberal suggestions] =&gt; Array
        (
            [slug] =&gt; liberal suggestions
            [parent] =&gt; 
            [name] =&gt; liberal suggestions
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; 
        )

    [licklider, j.c.r] =&gt; Array
        (
            [slug] =&gt; licklider, j.c.r
            [parent] =&gt; 
            [name] =&gt; Licklider, J.C.R.
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; 
        )

    [ligo] =&gt; Array
        (
            [slug] =&gt; ligo
            [parent] =&gt; 
            [name] =&gt; LiGO
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; LiGO {Linear Growth Operator} is a method to reduce the computation needed to train {{large language models} thus both improving computational effeiciency and reducing environmental cost.
        )

    [limb control] =&gt; Array
        (
            [slug] =&gt; limb control
            [parent] =&gt; 
            [name] =&gt; limb control
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; 
        )

    [linear discriminant] =&gt; Array
        (
            [slug] =&gt; linear discriminant
            [parent] =&gt; 
            [name] =&gt; linear discriminant
            [canonical] =&gt; linear discriminant analysis
            [see] =&gt; 
            [description] =&gt; 
        )

    [linear discriminant analysis] =&gt; Array
        (
            [slug] =&gt; linear discriminant analysis
            [parent] =&gt; 
            [name] =&gt; linear discriminant analysis
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; Linear discriminant analysis attemots to create one or more {{linear functions}} to separate known classes.  In the case of two classes and 2D data this is simply a line that runs between the classes.  Ideally the classes are completely separated by the linear functions, but in soem case this may be impossible, either becasue of noise blurring the boundaries (in which case the linear discrimant may still be a good classifier), or becasue the classes are not {{linearly separable}}.
        )

    [linear filter] =&gt; Array
        (
            [slug] =&gt; linear filter
            [parent] =&gt; 
            [name] =&gt; linear filter
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; Linear filters are functions/algorithms that work on numerical data including{{sequential data}}, such as {{time series}}, and {{images}}, but which only include linear combinations of the input.  Examples include compting a rolling seven-day average for daily data, or the {{Guassian filter}} for images. &lt;br&gt; &amp;nbsp; &amp;nbsp; &amp;nbsp; seven_day&lt;sub&gt;t&lt;/sub&gt; = (d&lt;sub&gt;t&lt;/sub&gt; + d&lt;sub&gt;t&amp;minus;1&lt;/sub&gt; + d&lt;sub&gt;t&amp;minus;2&lt;/sub&gt; + d&lt;sub&gt;t&amp;minus;3&lt;/sub&gt; + d&lt;sub&gt;t&amp;minus;4&lt;/sub&gt; + d&lt;sub&gt;t&amp;minus;5&lt;/sub&gt; + d&lt;sub&gt;t&amp;minus;6&lt;/sub&gt;) / 7
        )

    [linear filters] =&gt; Array
        (
            [slug] =&gt; linear filters
            [parent] =&gt; 
            [name] =&gt; linear filters
            [canonical] =&gt; linear filter
            [see] =&gt; 
            [description] =&gt; 
        )

    [linear filtering] =&gt; Array
        (
            [slug] =&gt; linear filtering
            [parent] =&gt; 
            [name] =&gt; linear filtering
            [canonical] =&gt; linear filter
            [see] =&gt; 
            [description] =&gt; 
        )

    [linear growth operator] =&gt; Array
        (
            [slug] =&gt; linear growth operator
            [parent] =&gt; 
            [name] =&gt; linear growth operator
            [canonical] =&gt; 
            [see] =&gt; ligo
            [description] =&gt; 
        )

    [linear inseparability] =&gt; Array
        (
            [slug] =&gt; linear inseparability
            [parent] =&gt; 
            [name] =&gt; linear inseparability
            [canonical] =&gt; linearly inseparable
            [see] =&gt; 
            [description] =&gt; 
        )

    [linearly inseparable] =&gt; Array
        (
            [slug] =&gt; linearly inseparable
            [parent] =&gt; 
            [name] =&gt; linearly inseparable
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; Two sets of data are linearly inseparable if there is no straight line (or {{hyperplane}} in higher dimensions) that spearates them. It is the opposite of {{linear separabile}}.
        )

    [linear patches] =&gt; Array
        (
            [slug] =&gt; linear patches
            [parent] =&gt; 
            [name] =&gt; linear patches
            [canonical] =&gt; 
            [see] =&gt; piecewise linear
            [description] =&gt; 
        )

    [linear planning] =&gt; Array
        (
            [slug] =&gt; linear planning
            [parent] =&gt; 
            [name] =&gt; linear planning
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; 
        )

    [linear regression] =&gt; Array
        (
            [slug] =&gt; linear regression
            [parent] =&gt; 
            [name] =&gt; linear regression
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; Linear regression is a {{statistical technique}} to fit a best line through 2D (x,y) data.   That is, it finds a line that predicts the y value ({{dependent variable}}) using the x value  ({{independent variable}}).  More formally, given data y&lt;sub&gt;i&lt;/sub&gt;, x&lt;sub&gt;i&lt;/sub&gt;, one is trying to find m and c to fit a line of the form y=mx+c.   The best values of m and c can be calculated using:&lt;br&gt; &amp;nbsp; &amp;nbsp; &amp;nbsp; m &amp;nbsp; = &amp;nbsp; &amp;sigma;&lt;sub&gt;x&lt;/sub&gt;&lt;sub&gt;y&lt;/sub&gt; / &amp;sigma;&lt;sup&gt;2&lt;/sup&gt;&lt;sub&gt;x&lt;/sub&gt;&lt;br&gt; &amp;nbsp; &amp;nbsp; &amp;nbsp; c &amp;nbsp; = &amp;nbsp; &amp;mu;&lt;sub&gt;y&lt;/sub&gt; &amp;minus; m &amp;mu;&lt;sub&gt;x&lt;/sub&gt;  &lt;br&gt; where &amp;mu;&lt;sub&gt;x&lt;/sub&gt; and &amp;mu;&lt;sub&gt;x&lt;/sub&gt; are the {{arithmetic mean}} of x and y respectively, &amp;sigma;&lt;sup&gt;2&lt;/sup&gt;&lt;sub&gt;x&lt;/sub&gt; is the {{variance}} of x, and  &amp;sigma;&lt;sub&gt;x&lt;/sub&gt;&lt;sub&gt;y&lt;/sub&gt; is the covariance of x and y&lt;br&gt; Note that the idea of 'best' here means minimising the sum of the squares of the y error, that is minimising: &lt;br&gt; &amp;nbsp; &amp;nbsp; &amp;nbsp; &amp;Sigma; (y&lt;sub&gt;i&lt;/sub&gt; &amp;minus; mx&lt;sub&gt;i&lt;/sub&gt;+c)&lt;sup&gt;2&lt;/sup&gt; &lt;br&gt; You get different answer if you swop the variables and look for the regression of x on y.  &lt;br&gt;&lt;br&gt; When there is more than one {{independent variable}}, there is an extension, {{multi-linear regression}}, which instead fits a {{hyperplane}}) to the data.
        )

    [linear separability] =&gt; Array
        (
            [slug] =&gt; linear separability
            [parent] =&gt; 
            [name] =&gt; linear separability
            [canonical] =&gt; linearly separable
            [see] =&gt; 
            [description] =&gt; 
        )

    [linearly separable] =&gt; Array
        (
            [slug] =&gt; linearly separable
            [parent] =&gt; 
            [name] =&gt; linearly separable
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; Two sets of data are linearly separable if there is a straight line (or {{hyperplane}} in higher dimensions) that separates them.  In other words there is a linear combination (a&lt;sub&gt;i&lt;/sub&gt;) of the features and a {{threshold}} T such that fp&lt;br&gt; &amp;nbsp;   &amp;nbsp;   &amp;nbsp;   a&lt;sub&gt;1&lt;/sub&gt;d&lt;sub&gt;1&lt;/sub&gt; +  a&lt;sub&gt;2&lt;/sub&gt;d&lt;sub&gt;2&lt;/sub&gt; + ...  + a&lt;sub&gt;n&lt;/sub&gt;d&lt;sub&gt;n&lt;/sub&gt;  is less than T for data items (d&lt;sub&gt;i&lt;/sub&gt;) in the first datasets and greater than T for the second &lt;br&gt; In such cases, the {{hyperplane}} acts as a {[classifier}}.
        )

    [link structure] =&gt; Array
        (
            [slug] =&gt; link structure
            [parent] =&gt; 
            [name] =&gt; link structure
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; 
        )

    [lisp] =&gt; Array
        (
            [slug] =&gt; lisp
            [parent] =&gt; 
            [name] =&gt; Lisp
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; Lisp is one of the most significant early AI programming langauges, known for its heavy use of brackets.  Lisp stands for &quot;List processing&quot; and the language is heavily influenced by the {{lambda calculus}}.  Lisp has several dialects including {{Scheme}}.  Lisp is an early {{functional programmon g}} lanugage although not a {{pure finctional language}} as it has updatabale variables and state.  
        )

    [llama] =&gt; Array
        (
            [slug] =&gt; llama
            [parent] =&gt; 
            [name] =&gt; LLaMA
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; LLaMA (large language model meta AI) is a {{foundation model}} released by {{Meta}}.  It was designed to be cheaper to initially train and fine tune than other {{large language models}} at the time.  The second iteration, LLaMA 2 is open source and this has allowed substantial further innovation.
        )

    [llama 2] =&gt; Array
        (
            [slug] =&gt; llama 2
            [parent] =&gt; 
            [name] =&gt; LLaMA 2
            [canonical] =&gt; llama
            [see] =&gt; 
            [description] =&gt; 
        )

    [loans example] =&gt; Array
        (
            [slug] =&gt; loans example
            [parent] =&gt; 
            [name] =&gt; loans example
            [canonical] =&gt; 
            [see] =&gt; production system
            [description] =&gt; This is an example {{production system}} about loan approvals used in  the AI textbook.  The example is coded in {{Prolog}} in the code examples for the book.
        )

    [local data access] =&gt; Array
        (
            [slug] =&gt; local data access
            [parent] =&gt; 
            [name] =&gt; local data access
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; Local data access is when the data required for code to execute is physiacally close to it, in the sense of on the same proessor, or at least close network.  This is an issue for {{big data}} processing especially on {{cloud computing}} platforms, where data may be stored in multiple computers often in physically remote {{data centres}}.  The goal is to perfom as much procesisng locally without moving data around.   Where the code is accessing ultiple data items, it is important that these are then stored physically close to one another.  This may be because theer are interincis commonlaity about data so they can be permamnently stored close to one another, or brought together for the purposes of processing. An exmaple all the former is data about a single person, and maybe summary inorrmation about their friends, in a {{social network}} which can be stored as a single record or data item.  An example of the latter is {{MapReduce}}, which uses a common {{hash}} value to bring items that need to be processed together by the {{reduce}} step on the same processor.
        )

    [local maxima] =&gt; Array
        (
            [slug] =&gt; local maxima
            [parent] =&gt; 
            [name] =&gt; local maxima
            [canonical] =&gt; local maximum
            [see] =&gt; 
            [description] =&gt; 
        )

    [local maximum] =&gt; Array
        (
            [slug] =&gt; local maximum
            [parent] =&gt; 
            [name] =&gt; local maximum
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; A local maximum for a function is a set of parameters for the function that is the largest in a neighbourhood, that is f(p) &amp;gt; f(p+&amp;delta;) for small changes $delta; in the parametre value.  In a physical landscape this would correspond to the top of a small hill, but where there may be a much larger hill elsewhere.   This geographic analogy can be helpful in the case of the {{firness landcape}} egnerated by a {{fitness function}} for {{machine learning}}; however, whlst the space of parametres for a physicla lcation is 2D (latitued,longitude), the space of parameters of a {{ftness functoon}} is often {{multi-dimensional}} and may include a mix of different kinds of parameters, not just numeric.   It is the opposite of a {{local minimum}} and is to be contrasted with a {{global maximum}} which is the largest/highest point overall.
        )

    [local minima] =&gt; Array
        (
            [slug] =&gt; local minima
            [parent] =&gt; 
            [name] =&gt; local minima
            [canonical] =&gt; local minimum
            [see] =&gt; 
            [description] =&gt; 
        )

    [local minima/maxima] =&gt; Array
        (
            [slug] =&gt; local minima/maxima
            [parent] =&gt; 
            [name] =&gt; local minima/maxima
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; In a {[fitness landscape}} there may be many {{local minima}} and {{local maxima}} that is points the locally are smaller or larger than neoghbouring points, but that are not the smallest or bggest overall.
        )

    [local minimum] =&gt; Array
        (
            [slug] =&gt; local minimum
            [parent] =&gt; 
            [name] =&gt; local minimum
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; A local minimum for a function is a set of parameters for the function that is the smallest in a neighbourhood, that is f(p) &amp;lt; f(p+&amp;delta;) for small changes $delta; in the parameter value.  In a physical landscape this would correspond to the bottom of a small hole, but where there may be a much deeper gulf.   It is the opposite of a {{local maximum}} and is to be contrasted with a {{global mimumum}} which is the smallest/lowest point overall.
        )

    [local planning] =&gt; Array
        (
            [slug] =&gt; local planning
            [parent] =&gt; 
            [name] =&gt; local planning
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; Local planning is about dealing with small scale navigation including {{obstacle avoidance}} or navigating through doorways or corridors.  It s in contrast with {{global planning}} which is about finding the best overall route, but ignores small details. 
        )

    [locality] =&gt; Array
        (
            [slug] =&gt; locality
            [parent] =&gt; 
            [name] =&gt; locality
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; We say an algorithm ir problem to be solved has locality, whe the date requred for each step is close to one another in memory.  This depends on the way in whch data is stored in memory, so that a good choice of {{knowledge representation}} can make algoritms ore effeicient.  However, some problems, particualy those involving complex {{graph processong}}, are intrinsically non-local -- no representation can entirely avod dusant memory accesses. This is particularly important when dealing with {{big data}} distributed over many storage devices or even, in {{cloud computing}} multiple {{data centres}} as non-local data access may involve substantial network traffic and delays waiting for it.  However, it can also be important in {{parallel processing}} on {{GPU}} chips, or improving cache performance  on single computers. 
        )

    [logarithm base] =&gt; Array
        (
            [slug] =&gt; logarithm base
            [parent] =&gt; 
            [name] =&gt; logarithm base
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; 
        )

    [logarithm bases] =&gt; Array
        (
            [slug] =&gt; logarithm bases
            [parent] =&gt; 
            [name] =&gt; logarithm bases
            [canonical] =&gt; logarithm base
            [see] =&gt; 
            [description] =&gt; 
        )

    [logarithmic transform] =&gt; Array
        (
            [slug] =&gt; logarithmic transform
            [parent] =&gt; 
            [name] =&gt; logarithmic transform
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; Taking the logorithm of data items can sometimes be a useful form of data transformation in several circumstances.  A logorithmic transform turns exponential growth onto linear grpwth, so can be useful for long term financial data where value of money gradually reduces in value due to inflation.  It can also be used where the variability grows with the size of data items, which is freqently the case for certain kinds of time measurements.  If you are measuring time delays often larger values also have larger variability.  If the logarithm of data follows the {{Normal distribution}}, the data is known as {{log-normal}}.
        )

    [logarithmic transformation] =&gt; Array
        (
            [slug] =&gt; logarithmic transformation
            [parent] =&gt; 
            [name] =&gt; logarithmic transformation
            [canonical] =&gt; logarithmic transform
            [see] =&gt; 
            [description] =&gt; 
        )

    [logic] =&gt; Array
        (
            [slug] =&gt; logic
            [parent] =&gt; 
            [name] =&gt; logic
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; 
            [subentries] =&gt; Array
                (
                    [logic!as knowledge representation] =&gt; Array
                        (
                            [slug] =&gt; logic!as knowledge representation
                            [parent] =&gt; logic
                            [name] =&gt; as knowledge representation
                            [canonical] =&gt; 
                            [see] =&gt; 
                            [description] =&gt; 
                        )

                    [logic!fuzzy] =&gt; Array
                        (
                            [slug] =&gt; logic!fuzzy
                            [parent] =&gt; logic
                            [name] =&gt; fuzzy
                            [canonical] =&gt; 
                            [see] =&gt; fuzzy logic
                            [description] =&gt; 
                        )

                    [logic!operators] =&gt; Array
                        (
                            [slug] =&gt; logic!operators
                            [parent] =&gt; logic
                            [name] =&gt; operators
                            [canonical] =&gt; 
                            [see] =&gt; 
                            [description] =&gt; 
                        )

                    [logic!standard] =&gt; Array
                        (
                            [slug] =&gt; logic!standard
                            [parent] =&gt; logic
                            [name] =&gt; standard
                            [canonical] =&gt; 
                            [see] =&gt; standard logic
                            [description] =&gt; 
                        )

                )

        )

    [logic!as knowledge representation] =&gt; Array
        (
            [slug] =&gt; logic!as knowledge representation
            [parent] =&gt; logic
            [name] =&gt; as knowledge representation
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; 
        )

    [logic!fuzzy] =&gt; Array
        (
            [slug] =&gt; logic!fuzzy
            [parent] =&gt; logic
            [name] =&gt; fuzzy
            [canonical] =&gt; 
            [see] =&gt; fuzzy logic
            [description] =&gt; 
        )

    [logic!operators] =&gt; Array
        (
            [slug] =&gt; logic!operators
            [parent] =&gt; logic
            [name] =&gt; operators
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; 
        )

    [logic!standard] =&gt; Array
        (
            [slug] =&gt; logic!standard
            [parent] =&gt; logic
            [name] =&gt; standard
            [canonical] =&gt; 
            [see] =&gt; standard logic
            [description] =&gt; 
        )

    [logistic function] =&gt; Array
        (
            [slug] =&gt; logistic function
            [parent] =&gt; 
            [name] =&gt; logistic function
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; A logistic function is a form of {{sigmoid functiion}} often used as a weak {[threshold}} in an {{neural network}}.  Like a {{step function}} it maps unbounded values to a finite [0,1] range and is zero for large negatuve values and one for lareg postuve values.  However, unlike a  {{step function}} the prcess is smooth with a near {{linear}} section in the middle and gently asymptotes for more edxtreme values.  This is imprtant for {{machine learning}} as, in general, it is easier to learn {[contnuous}} features. 
        )

    [long-tail distribution] =&gt; Array
        (
            [slug] =&gt; long-tail distribution
            [parent] =&gt; 
            [name] =&gt; long-tail distribution
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; A long-tail distribution is a form of {{probability distribution}} or empirical histogram of values, where the number of extremely large elements may be sufficient to make a major difference to common measures such as the {{mean}} or {[standard deviation}}.  A common example of a long-tailed distribution is the {{power-law distrubution}} often encounterd in {{network data}} such as  {{social media}} connections or {{web links}}.  Income distrubutions in most soieties are less extreme than the power law, and usually has a well-defined mean but still have effectively unbounded {{variance}}.  Note that some distributions such as the {{Normal distribution}} may have a long tail in the sense that there is the potential for a few very large values, but these are exteremely rare.
        )

    [long-tail distributions] =&gt; Array
        (
            [slug] =&gt; long-tail distributions
            [parent] =&gt; 
            [name] =&gt; long-tail distributions
            [canonical] =&gt; long-tail distribution
            [see] =&gt; 
            [description] =&gt; 
        )

    [long-term memory] =&gt; Array
        (
            [slug] =&gt; long-term memory
            [parent] =&gt; 
            [name] =&gt; long-term memory
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; Long-term memory in humans refers to our memory of events, people, ideas and skills that stay with us for very long periods, perhaps for our whole lives.  It is often divided into: {{episodic memory}} of events and gtheor relations; {[declarative memory}} of facts and concepts; and {{procedural memory}} of how to do things.  In addition, {{decalartuve memory}} for different kinds of things, for example perosnal relationships vs public figures, are stored in different ways.
        )

    [long-term potentiation] =&gt; Array
        (
            [slug] =&gt; long-term potentiation
            [parent] =&gt; 
            [name] =&gt; long-term potentiation
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; Long-term potentiation refers to the way in which chemcial changes in individual {{neurones}} can alter firing (either making it easier or harder) for a period of time from seconds to hours.  This can give rise to {{mezzanine memory}} or {{long-term working memory}} that lasts over periods of up to a few hours.  In contrast, human {{short term memory}} and {[conscious reasoning}} are deternines by very short term electrical (ionic activity) and {{long-term memory}} is based on actual growth or decay of the {{axions}} and {{synapses}} that connect {{neurones}}.
        )

    [long-term short-term memory networks] =&gt; Array
        (
            [slug] =&gt; long-term short-term memory networks
            [parent] =&gt; 
            [name] =&gt; long-term short-term memory networks
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; Long-term short-term memory networks (LTSM) are used on {{time series data}}.  LTSM have special memory nodes that can be triggered to remember the state of their input based on the values of other parts of a {{neural network}}.  This can be used to model effects such as systems that have long phases of similar behaviour, with occasional changes to other kinds of behaviour, or to recall important events for long periods.
        )

    [long-term short-term network] =&gt; Array
        (
            [slug] =&gt; long-term short-term network
            [parent] =&gt; 
            [name] =&gt; long-term short-term network
            [canonical] =&gt; long-term short-term memory networks
            [see] =&gt; 
            [description] =&gt; 
        )

    [lora] =&gt; Array
        (
            [slug] =&gt; lora
            [parent] =&gt; 
            [name] =&gt; LoRA
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; LoRA (low-rank adaptation of large language models} performs {{dimensional reduction}} on inner layers of the {{deep neural network}} used in a {{large language model}} thus reducing the cost of execution and retraining.  This has economic and environmental advantages.
        )

    [lovelace, ada] =&gt; Array
        (
            [slug] =&gt; lovelace, ada
            [parent] =&gt; 
            [name] =&gt; Lovelace, Ada
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; Ada Lovelace was a collaborator with {{Charles Babbage}} in the 18th Century, and daughter of the poet Lord Byron.  She is regarded as the first programmer, designing algorithsm that would run on the {{Analytical Engine}},. 
 Critically also she was an early philosopher of computation, identifiying the mportance of &lt;em&gt;symbolic results&lt;/em&gt; and difference between calcuation and computation:&lt;br&gt; 
&quot;&lt;em&gt;The {{Difference Engine}} can merely tabulate, and is incapable of developing, the {{Analytical Engine}} can either tabulate or develope&lt;/em&gt;&quot;
&lt;br&gt;
She also wrpte in aesthetic terms appreciating the  &quot;&lt;em&gt;intrinsic beauty, symmetry and logical completeness&lt;\em&gt;&quot; of mathemtatcal sciences  and the way that the {{Analytical Engine}} &quot;&lt;em&gt;weaves algebraical patterns just as the {{Jacquard-loom}} weaves flowers and leaves&lt;/em&gt;&quot;.
        )

    [low-rank adaptation of large language models] =&gt; Array
        (
            [slug] =&gt; low-rank adaptation of large language models
            [parent] =&gt; 
            [name] =&gt; low-rank adaptation of large language models
            [canonical] =&gt; lora
            [see] =&gt; 
            [description] =&gt; 
        )

    [lower bound] =&gt; Array
        (
            [slug] =&gt; lower bound
            [parent] =&gt; 
            [name] =&gt; lower bound
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; A lower bound is an estimate, {{heuristic}} or calculated value that yoo know lies at or below the smallest possible value of an unknown value or function. For example, if you are looking for the road distance between two locations, you know this will be at least as large as the {{Euclidean dstace}} between them (crow flies distance).  Knowing a lower bound can sometimes allow whole branches of a {{search tree}} or similar structure to be {{pruned}}.  For example, if you are looking for the smallest value in a tree and have found a node with value 7, and then consider an alternative branch in the tree where a {{heuristic}} tells you that the {{lower bound}} for nodes in the branch is 12, you can safely ignore the entre sub-tree.
        )

    [ltsm] =&gt; Array
        (
            [slug] =&gt; ltsm
            [parent] =&gt; 
            [name] =&gt; LTSM
            [canonical] =&gt; long-term short-term memory networks
            [see] =&gt; 
            [description] =&gt; 
        )

    [machine learning] =&gt; Array
        (
            [slug] =&gt; machine learning
            [parent] =&gt; 
            [name] =&gt; machine learning
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; 
            [subentries] =&gt; Array
                (
                    [machine learning!application] =&gt; Array
                        (
                            [slug] =&gt; machine learning!application
                            [parent] =&gt; machine learning
                            [name] =&gt; application
                            [canonical] =&gt; 
                            [see] =&gt; 
                            [description] =&gt; 
                        )

                    [machine learning!training] =&gt; Array
                        (
                            [slug] =&gt; machine learning!training
                            [parent] =&gt; machine learning
                            [name] =&gt; training
                            [canonical] =&gt; 
                            [see] =&gt; 
                            [description] =&gt; 
                        )

                    [machine learning!validation] =&gt; Array
                        (
                            [slug] =&gt; machine learning!validation
                            [parent] =&gt; machine learning
                            [name] =&gt; validation
                            [canonical] =&gt; 
                            [see] =&gt; 
                            [description] =&gt; 
                        )

                )

        )

    [machine learning!application] =&gt; Array
        (
            [slug] =&gt; machine learning!application
            [parent] =&gt; machine learning
            [name] =&gt; application
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; 
        )

    [machine learning!training] =&gt; Array
        (
            [slug] =&gt; machine learning!training
            [parent] =&gt; machine learning
            [name] =&gt; training
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; 
        )

    [machine learning!validation] =&gt; Array
        (
            [slug] =&gt; machine learning!validation
            [parent] =&gt; machine learning
            [name] =&gt; validation
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; 
        )

    [machine-learning
] =&gt; Array
        (
            [slug] =&gt; machine-learning

            [parent] =&gt; 
            [name] =&gt; machine-learning

            [canonical] =&gt; machine learning
            [see] =&gt; 
            [description] =&gt; 
        )

    [macro languages] =&gt; Array
        (
            [slug] =&gt; macro languages
            [parent] =&gt; 
            [name] =&gt; macro languages
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; 
        )

    [magic square] =&gt; Array
        (
            [slug] =&gt; magic square
            [parent] =&gt; 
            [name] =&gt; magic square
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; A magic square is a square grid of numbers, typically consecutuve, where each row, column and sometimes diagnonal add up to the same number.
        )

    [magic squares] =&gt; Array
        (
            [slug] =&gt; magic squares
            [parent] =&gt; 
            [name] =&gt; magic squares
            [canonical] =&gt; 
            [see] =&gt; magic square
            [description] =&gt; 
        )

    [manhattan block distance] =&gt; Array
        (
            [slug] =&gt; manhattan block distance
            [parent] =&gt; 
            [name] =&gt; Manhatten block distance
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; The Manhatten block distance simply adds the x and y distance like walking the streets in a city.  In higher dimensions it is simply the sum of the absolute distance along every dimension, or in data terms the sum of the difference of every {{feature}}.  Alternative distance measures include the {{Euclidean distance}}, which is the 'as the crows flies' distance between two point
        )

    [map] =&gt; Array
        (
            [slug] =&gt; map
            [parent] =&gt; 
            [name] =&gt; map
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; Map is a standard function in {{Lisp}} and other {{functional programming}} languages.  Map has two arguments map(f,list). where the function f is applied to every element of the list forming a new list.  That is &lt;br&gt; map( f, (a,b,c...z) )  =  ( f(a), f(b), f(c), ... f(z) ).  &lt;br&gt;&lt;br&gt; The map function is one of the inspirations for the {{MapReduce}} infrastructure for massive scale processing of {{big data}}.
        )

    [map!lisp] =&gt; Array
        (
            [slug] =&gt; map!lisp
            [parent] =&gt; map
            [name] =&gt; lisp
            [canonical] =&gt; map
            [see] =&gt; 
            [description] =&gt; 
        )

    [mapreduce] =&gt; Array
        (
            [slug] =&gt; mapreduce
            [parent] =&gt; 
            [name] =&gt; MapReduce
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; MapReduce is a distributed computing framework for delaing with {{big data}}.  It is inspired by the {{map}} and {{reduce}} functions found in {{Lisp}} and other {{functional programming}} languages: the {{map}} stage initially processes each item of data separately and then the {[reduce}} stage collects and aggregates related intermediate results.  However, MapReduce has two crucial  additional features : (i) the use of a hash as means to distribute intermediate results over different computers, so redcuing the likelihood of {{Byzantine conditions}}; and (ii) means to monitor for failure of individual computers and recvover from this.    &lt;br&gt;&lt;table align=&quot;center&quot;&gt;   &lt;tr&gt;&lt;td&gt;each data item&lt;/td&gt;&lt;td&gt;&amp;rarr; &lt;/td&gt;&lt;td align=&quot;center&quot;&gt;map &lt;/td&gt;&lt;td&gt;&amp;rarr; &lt;/td&gt; &lt;td&gt;hash + processed data&lt;/td&gt;&lt;/tr&gt;    &lt;tr&gt;&lt;td&gt;all data for a hash&lt;/td&gt;&lt;td&gt;&amp;rarr; &lt;/td&gt;&lt;td align=&quot;center&quot;&gt;reduce &lt;/td&gt;&lt;td&gt;&amp;rarr; &lt;/td&gt; &lt;td&gt;one or more agggregated calculations for the hashed data &lt;br&gt;(possibly including fresh has for further reduce steps)&lt;/td&gt;&lt;/tr&gt;    &lt;/table&gt;.    &lt;br&gt;MapReduce was initially developed by engineers at Google, but has since become part of the open source Apache {{Hadoop}} project.  
        )

    [markov, andrey] =&gt; Array
        (
            [slug] =&gt; markov, andrey
            [parent] =&gt; 
            [name] =&gt; Markov, Andrey
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; Andrey Andreyevich Markov was a Russian mathematician, especially well known for the concept of a {{Markov model}} or {{Markov process}}
        )

    [markov model] =&gt; Array
        (
            [slug] =&gt; markov model
            [parent] =&gt; 
            [name] =&gt; Markov model
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; A Markov model is a {{stochastic process}} where the probability of each step in the sequence is dependent only on the previous step, or in the case of {{higher-order Markov models}} the last n-steps.  Markov models are frequently shown as matrices where each entry in the matrix M&lt;sub&gt;ik&lt;/sub&gt; represents the {{conditional probability}} that if the current state is i, then the next state will be k.
        )

    [marr's primal sketch] =&gt; Array
        (
            [slug] =&gt; marr's primal sketch
            [parent] =&gt; 
            [name] =&gt; Marr's primal sketch
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; 
        )

    [mary shelley] =&gt; Array
        (
            [slug] =&gt; mary shelley
            [parent] =&gt; 
            [name] =&gt; Mary Shelley
            [canonical] =&gt; 
            [see] =&gt; shelley, mary
            [description] =&gt; 
        )

    [materialism] =&gt; Array
        (
            [slug] =&gt; materialism
            [parent] =&gt; 
            [name] =&gt; materialism
            [canonical] =&gt; 
            [see] =&gt; physicalism
            [description] =&gt; 
        )

    [mds] =&gt; Array
        (
            [slug] =&gt; mds
            [parent] =&gt; 
            [name] =&gt; MDS
            [canonical] =&gt; 
            [see] =&gt; multi-dimensional scaling
            [description] =&gt; 
        )

    [means-ends analysis] =&gt; Array
        (
            [slug] =&gt; means-ends analysis
            [parent] =&gt; 
            [name] =&gt; means-ends analysis
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; 
        )

    [memorizing] =&gt; Array
        (
            [slug] =&gt; memorizing
            [parent] =&gt; 
            [name] =&gt; memorizing
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; 
        )

    [memory] =&gt; Array
        (
            [slug] =&gt; memory
            [parent] =&gt; 
            [name] =&gt; memory
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; 
        )

    [memristor] =&gt; Array
        (
            [slug] =&gt; memristor
            [parent] =&gt; 
            [name] =&gt; memristor
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; 
        )

    [method acting] =&gt; Array
        (
            [slug] =&gt; method acting
            [parent] =&gt; 
            [name] =&gt; method acting
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; 
        )

    [microdata] =&gt; Array
        (
            [slug] =&gt; microdata
            [parent] =&gt; 
            [name] =&gt; microdata
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; 
        )

    [microformats] =&gt; Array
        (
            [slug] =&gt; microformats
            [parent] =&gt; 
            [name] =&gt; microformats
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; Microformats are ways to encode {{semantic knowledge}} into web pages.  Special classes are used around elements, such as dates, personal details, etc.  For example: &lt;br&gt; &amp;lt;span class=&quot;p-name&quot;&amp;gt;Alan Dix&amp;lt/span&amp;gt &lt;br&gt; This means that a computer reading the page can interpret the marked elements, hence turning the human-readable web page into machine-readable data.
        )

    [microtubules] =&gt; Array
        (
            [slug] =&gt; microtubules
            [parent] =&gt; 
            [name] =&gt; microtubules
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; Microtubules are tiny structures within biological cells that act rather like scaffolding.  They are what enable single celled organisms, such as amoeba, to move and also pull chromosomes apart and re-assemble them during cell division.  {{Roger Penrose}} has suggested that {{quantum effects}} within microtubules may be related to {{human intelligence}}.
        )

    [minimax] =&gt; Array
        (
            [slug] =&gt; minimax
            [parent] =&gt; 
            [name] =&gt; minimax
            [canonical] =&gt; minimax search
            [see] =&gt; 
            [description] =&gt; 
        )

    [minimax algorithm] =&gt; Array
        (
            [slug] =&gt; minimax algorithm
            [parent] =&gt; 
            [name] =&gt; minimax algorithm
            [canonical] =&gt; minimax search
            [see] =&gt; 
            [description] =&gt; 
        )

    [minimax search] =&gt; Array
        (
            [slug] =&gt; minimax search
            [parent] =&gt; 
            [name] =&gt; minimax search
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; Minimax is a search algorithm for playing {{zero-sum games}}.  At each stage, the player whose turn it is looks ahead at each possible move and works out the maximum score the opponent could get if it plays optimally.  The player then chooses the move that minimises the opponents score and hence maxmises their own.  The same algorithm is applied for working out the opponent's move down each branch and so on.
        )

    [misinformation] =&gt; Array
        (
            [slug] =&gt; misinformation
            [parent] =&gt; 
            [name] =&gt; misinformation
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; 
        )

    [modal] =&gt; Array
        (
            [slug] =&gt; modal
            [parent] =&gt; 
            [name] =&gt; modal
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; 
        )

    [mode] =&gt; Array
        (
            [slug] =&gt; mode
            [parent] =&gt; 
            [name] =&gt; mode
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; 
        )

    [monotonic reasoning] =&gt; Array
        (
            [slug] =&gt; monotonic reasoning
            [parent] =&gt; 
            [name] =&gt; monotonic reasoning
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; 
        )

    [monte carlo algorithm] =&gt; Array
        (
            [slug] =&gt; monte carlo algorithm
            [parent] =&gt; 
            [name] =&gt; monte carlo algorithm
            [canonical] =&gt; 
            [see] =&gt; monte carlo techniques
            [description] =&gt; 
        )

    [monte carlo search] =&gt; Array
        (
            [slug] =&gt; monte carlo search
            [parent] =&gt; 
            [name] =&gt; Monte Carlo search
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; A Monte Carlo search uses random choices as part of the search algorithm.  An extreme case is random {{generate and test}}, choosing points in the {{search space}} at random until one satisfies the search criterion.   However, forms of Monte Carlo search can also use {{heuristics}} to nudge the algorothm towards more likely parts of the {{search space}}.
        )

    [monte carlo techniques] =&gt; Array
        (
            [slug] =&gt; monte carlo techniques
            [parent] =&gt; 
            [name] =&gt; Monte Carlo techniques
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; Monte Carlo techniques use some form of random choices as part of their algorithms.  Examples include {{random forests}} and {{Monte Carlo search}}.  These can be practical algorithms, especially where the choice is not entirely random, but guided by {{heuristics}} to enable different chocies to have different probabilities.   Monto Carlo algroithms can also be used as a base case to compare with other algorithms.  For example, a simple Monte Carlo {{optimisation}} algorithm is to simply choose parameters at random and keep track of the best seen so far; if other algorithms do not perform significantly better than this on a data set, then they are clearly not well suited to the kind of data.
        )

    [monte carlo tree search] =&gt; Array
        (
            [slug] =&gt; monte carlo tree search
            [parent] =&gt; 
            [name] =&gt; Monte Carlo tree search
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; Monte Carlo tree search is any form of {{Monte Carlo search}} applied to trees.  For example a simple variant of depth search could randomly choose branches to follow, stopping when the path gets to a leaf, or, in the case of very large trees, at a random point before (getting bored!).  As with any {{Monte Carlo algorithm}}, heuristics can be used to assign probabilities to different branches. {{AlphaGo}} used {{Monte Carlo  tree search}} alongside {{heuristics}} generated by {{deep neuaral networks}}.
        )

    [moving average] =&gt; Array
        (
            [slug] =&gt; moving average
            [parent] =&gt; 
            [name] =&gt; moving average
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; 
        )

    [moving average models] =&gt; Array
        (
            [slug] =&gt; moving average models
            [parent] =&gt; 
            [name] =&gt; moving average models
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; 
        )

    [moving images] =&gt; Array
        (
            [slug] =&gt; moving images
            [parent] =&gt; 
            [name] =&gt; moving images
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; 
        )

    [moving pictures] =&gt; Array
        (
            [slug] =&gt; moving pictures
            [parent] =&gt; 
            [name] =&gt; moving pictures
            [canonical] =&gt; 
            [see] =&gt; moving images
            [description] =&gt; 
        )

    [multi-dimensional scaling] =&gt; Array
        (
            [slug] =&gt; multi-dimensional scaling
            [parent] =&gt; 
            [name] =&gt; multi-dimensional scaling
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; Multi-dimensional scaling (MDS) creates a two (usually) dimensioanl represention of data that preserves closeness.  Its input is a {{similarity matrix}} (S&lt;sub&gt;ik&lt;/sub&gt;) or a  {{dissimilarity matrix}} (D&lt;sub&gt;ik&lt;/sub&gt;)  between items an then allocates for each item i coordinates (x&lt;sub&gt;i&lt;/sub&gt;,y&lt;sub&gt;i&lt;/sub&gt;) such that, on average, when the {{Euclidean distance}} between (x&lt;sub&gt;i&lt;/sub&gt;,y&lt;sub&gt;i&lt;/sub&gt;)  and (x&lt;sub&gt;j&lt;/sub&gt;,y&lt;sub&gt;j&lt;/sub&gt;) is small then S&lt;sub&gt;ik&lt;/sub&gt;) is large (or D&lt;sub&gt;ik&lt;/sub&gt;) is small) and vice versa.
        )

    [multi-layer perceptron] =&gt; Array
        (
            [slug] =&gt; multi-layer perceptron
            [parent] =&gt; 
            [name] =&gt; multi-layer perceptron
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; 
        )

    [multi-linear regression] =&gt; Array
        (
            [slug] =&gt; multi-linear regression
            [parent] =&gt; 
            [name] =&gt; multi-linear regression
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; Multi-objective optimisation is used when there are sevral criteri, all of which one would wich to optimise.  For example, a company may want to optimise profit, welfare for its workers and environmental impact.  Typically there is not a way to simultaneously optimise all the criteria so there has to be a trade-off between them. Multi-objective optimisation is hence far more difficult than simple single-criterion {{optimisation}}. One solution is to weight the criteria and optmise the weighted sum.  Other approaches attempt to find some or all {{pareto-optimal}} solutions; that is solutions from which there is no way to improve all citeria.  
        )

    [multi-objective optimisation] =&gt; Array
        (
            [slug] =&gt; multi-objective optimisation
            [parent] =&gt; 
            [name] =&gt; multi-objective optimisation
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; 
        )

    [multimodel interfaces] =&gt; Array
        (
            [slug] =&gt; multimodel interfaces
            [parent] =&gt; 
            [name] =&gt; multimodel interfaces
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; Multimodel interfaces use several human modes of interaction simultaneously ; for example speaking whilst pointing at a screen and saying &quot;put that there&quot;.
        )

    [multiple images] =&gt; Array
        (
            [slug] =&gt; multiple images
            [parent] =&gt; 
            [name] =&gt; multiple images
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; 
        )

    [mutual information] =&gt; Array
        (
            [slug] =&gt; mutual information
            [parent] =&gt; 
            [name] =&gt; mutual information
            [canonical] =&gt; 
            [see] =&gt; information gain
            [description] =&gt; 
        )

    [mutual inhibition] =&gt; Array
        (
            [slug] =&gt; mutual inhibition
            [parent] =&gt; 
            [name] =&gt; mutual inhibition
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; 
        )

    [mycin] =&gt; Array
        (
            [slug] =&gt; mycin
            [parent] =&gt; 
            [name] =&gt; MYCIN
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; 
        )

    [mycin!rule] =&gt; Array
        (
            [slug] =&gt; mycin!rule
            [parent] =&gt; mycin
            [name] =&gt; rule
            [canonical] =&gt; mycin
            [see] =&gt; 
            [description] =&gt; 
        )

    [mycin rule] =&gt; Array
        (
            [slug] =&gt; mycin rule
            [parent] =&gt; 
            [name] =&gt; mycin rule
            [canonical] =&gt; mycin
            [see] =&gt; 
            [description] =&gt; 
        )

    [n-gram] =&gt; Array
        (
            [slug] =&gt; n-gram
            [parent] =&gt; 
            [name] =&gt; n-gram
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; In a text or sequence an n-gram is a sequence of n consecutive tokens.   In the case of text treated as a sequence words, a 1-gram is a single word in the text and a 2-gram is a word pair. For example, in this descriotion, &quot;text or sequence a&quot; is a 4-gram.  Typically you are interested in the set of all n-grams in a text  for some n and their respective frequencies of occurrence in the text.
        )

    [named entity recognition] =&gt; Array
        (
            [slug] =&gt; named entity recognition
            [parent] =&gt; 
            [name] =&gt; named entity recognition
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; Named entity recognition is a form of {{entity recognition}} where the entities being detected have a proper name such a person's name or a placename.  Whilst datatyopes such as numbers and date may use some ofrm of {{pattern matching}} named entity recognition usually requires looking up individual words or {{n-grams}} in large lexicons or personal data sources.
        )

    [named graphs] =&gt; Array
        (
            [slug] =&gt; named graphs
            [parent] =&gt; 
            [name] =&gt; named graphs
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; 
        )

    [namespaces] =&gt; Array
        (
            [slug] =&gt; namespaces
            [parent] =&gt; 
            [name] =&gt; namespaces
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; 
        )

    [national insurance number] =&gt; Array
        (
            [slug] =&gt; national insurance number
            [parent] =&gt; 
            [name] =&gt; National Insurance number
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; The National Insurance number (or NI number) is the principal {{personal identifier}} used in the UK for government purposes.  There are separate identifiers for health and tax.
        )

    [natural language database query] =&gt; Array
        (
            [slug] =&gt; natural language database query
            [parent] =&gt; 
            [name] =&gt; natural language database query
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; 
        )

    [natural language processing] =&gt; Array
        (
            [slug] =&gt; natural language processing
            [parent] =&gt; 
            [name] =&gt; natural language processing
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; 
            [subentries] =&gt; Array
                (
                    [natural language processing!database access] =&gt; Array
                        (
                            [slug] =&gt; natural language processing!database access
                            [parent] =&gt; natural language processing
                            [name] =&gt; database access
                            [canonical] =&gt; 
                            [see] =&gt; natural language database query
                            [description] =&gt; 
                        )

                    [natural language processing!database queries] =&gt; Array
                        (
                            [slug] =&gt; natural language processing!database queries
                            [parent] =&gt; natural language processing
                            [name] =&gt; database queries
                            [canonical] =&gt; 
                            [see] =&gt; natural language database query
                            [description] =&gt; 
                        )

                    [natural language processing!information management] =&gt; Array
                        (
                            [slug] =&gt; natural language processing!information management
                            [parent] =&gt; natural language processing
                            [name] =&gt; information management
                            [canonical] =&gt; 
                            [see] =&gt; 
                            [description] =&gt; 
                        )

                    [natural language processing!interfaces] =&gt; Array
                        (
                            [slug] =&gt; natural language processing!interfaces
                            [parent] =&gt; natural language processing
                            [name] =&gt; interfaces
                            [canonical] =&gt; 
                            [see] =&gt; 
                            [description] =&gt; 
                        )

                    [natural language processing!tagger] =&gt; Array
                        (
                            [slug] =&gt; natural language processing!tagger
                            [parent] =&gt; natural language processing
                            [name] =&gt; tagger
                            [canonical] =&gt; 
                            [see] =&gt; 
                            [description] =&gt; 
                        )

                    [natural language processing!tagset] =&gt; Array
                        (
                            [slug] =&gt; natural language processing!tagset
                            [parent] =&gt; natural language processing
                            [name] =&gt; tagset
                            [canonical] =&gt; 
                            [see] =&gt; 
                            [description] =&gt; 
                        )

                )

        )

    [natural language processing!database access] =&gt; Array
        (
            [slug] =&gt; natural language processing!database access
            [parent] =&gt; natural language processing
            [name] =&gt; database access
            [canonical] =&gt; 
            [see] =&gt; natural language database query
            [description] =&gt; 
        )

    [natural language processing!database queries] =&gt; Array
        (
            [slug] =&gt; natural language processing!database queries
            [parent] =&gt; natural language processing
            [name] =&gt; database queries
            [canonical] =&gt; 
            [see] =&gt; natural language database query
            [description] =&gt; 
        )

    [natural language processing!information management] =&gt; Array
        (
            [slug] =&gt; natural language processing!information management
            [parent] =&gt; natural language processing
            [name] =&gt; information management
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; 
        )

    [natural language processing!interfaces] =&gt; Array
        (
            [slug] =&gt; natural language processing!interfaces
            [parent] =&gt; natural language processing
            [name] =&gt; interfaces
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; 
        )

    [natural language processing!tagger] =&gt; Array
        (
            [slug] =&gt; natural language processing!tagger
            [parent] =&gt; natural language processing
            [name] =&gt; tagger
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; 
        )

    [natural language processing!tagset] =&gt; Array
        (
            [slug] =&gt; natural language processing!tagset
            [parent] =&gt; natural language processing
            [name] =&gt; tagset
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; 
        )

    [natural language understanding] =&gt; Array
        (
            [slug] =&gt; natural language understanding
            [parent] =&gt; 
            [name] =&gt; natural language understanding
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; 
        )

    [natural logarithm] =&gt; Array
        (
            [slug] =&gt; natural logarithm
            [parent] =&gt; 
            [name] =&gt; natural logarithm
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; The natural logarithm is a {{logarithm}} using e (Euler's number ~ 2.72) as base, it is often denoted &quot;ln&quot;.  That is ln(x) is the natural logarithm of x, and x = e&lt;sup&gt;ln(x)&lt;/sup&gt;.  Other common {{logarithm bases}} are 2 and 10, which are easier to interpret.  However natural logairithms occur in many mathematical and physical processes.
        )

    [natural selection] =&gt; Array
        (
            [slug] =&gt; natural selection
            [parent] =&gt; 
            [name] =&gt; natural selection
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; In biology {{natural selection}} refers to the way that indviduals who are more fit or able to cope in their envirnment are more likely to survive to breeding age and bring up offspring successfully ({{survival of the fitest}}).  This means that their {[genes}} propogate more to the next generation leading to a gradual shift of the population towards a more successful gene pool for the enviornment.  In the natural environment this process can happen relatvely quickly, espeically in fast breeding organisms such as insects or disease pathogens.  It is one of the processes behind {{Darwinian evolution}}, although the latter also requires genetic {{mutation}} to allow the gene pool to gradually shift also.  These processes are emulated in artifical {{genetic algorithms}}.
        )

    [negation as failure] =&gt; Array
        (
            [slug] =&gt; negation as failure
            [parent] =&gt; 
            [name] =&gt; negation as failure
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; Some implemetations of logic, notably {{Prolog}} apply negation as failure, whuch means that if a solution is not found (failure), the thing you are trying to solve is deemed as not true (negation).  This effectvely adopts the {{closed world assumption}} - if raw facts are not explciitly stated they are false, or in other words all facts are known in the {{knowledge base}}.
        )

    [nettalk] =&gt; Array
        (
            [slug] =&gt; nettalk
            [parent] =&gt; 
            [name] =&gt; NETtalk 
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; NETtalk was an early text to speech application built using a {{multi-layer perceptron}} built in the mid 1980s..  In some way NETtalk's training is very smilar to more recent systems, using data form surrounding letters to predict the ecorrect phoneme to pronounce.  However, in comparison the the many millions of neurons in current networks, NETtalk had a mere  203 input units, 80 hidden units and 26 output units. Yet despite this produced intelligable output and learning behavour rather like a child's.   Certainly a lesson for those today trying to reduce the {{carbon footprint}} of AI and reduce the {{digital divide}}.
        )

    [network representation] =&gt; Array
        (
            [slug] =&gt; network representation
            [parent] =&gt; 
            [name] =&gt; network representation
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; 
        )

    [network representations] =&gt; Array
        (
            [slug] =&gt; network representations
            [parent] =&gt; 
            [name] =&gt; network representations
            [canonical] =&gt; network representation
            [see] =&gt; 
            [description] =&gt; 
        )

    [neural correlates of consciousness] =&gt; Array
        (
            [slug] =&gt; neural correlates of consciousness
            [parent] =&gt; 
            [name] =&gt; neural correlates of consciousness
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; Neural correlates of consciousness are processes and behaviours in the brain that occur at the same time as {{subjective}} {{conscious}} experience, for example, {{global synchronisation of oscillations}}  where distant parts of the brain fire in the same rhythm.  As the maxim goies &quot;{{correlation does not imply causation}}&quot;, so one cannot say that the experince consciousness .is because of the particular neural phenomenon, but it does help to sfta nd inspire more theoretical, philosophical and practical approaches to {{consciousness}}.
        )

    [neural network] =&gt; Array
        (
            [slug] =&gt; neural network
            [parent] =&gt; 
            [name] =&gt; neural network
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; 
        )

    [neural networks] =&gt; Array
        (
            [slug] =&gt; neural networks
            [parent] =&gt; 
            [name] =&gt; neural networks
            [canonical] =&gt; neural network
            [see] =&gt; 
            [description] =&gt; 
        )

    [neuron] =&gt; Array
        (
            [slug] =&gt; neuron
            [parent] =&gt; 
            [name] =&gt; neuron
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; 
        )

    [neurones] =&gt; Array
        (
            [slug] =&gt; neurones
            [parent] =&gt; 
            [name] =&gt; neurones
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; Neurone is an alternative spelling of {{neuron}}.  Sometimes neurone, with the 'e' may be used to emphasise when the subjcet is bilogical rather than artificoal neurons.
        )

    [neurosymbolc ai] =&gt; Array
        (
            [slug] =&gt; neurosymbolc ai
            [parent] =&gt; 
            [name] =&gt; neurosymbolc AI
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; Neurosymbolc AI refers to techniques that combine {{sub-symbolic}} {{neural networks}} with {{symbolic}} techniques.  These may be fairly distinct modules in a {{hybrid archtecture}}, but also can be integrated at.a more fundamental way such as using neural techniques to reason over {{semantic networks}} or the way high-level {{counterfactual reasonng}} is used to guide low-level stimulus-response learning in the {{cognitive model of regret}}.
        )

    [neurosymbolic] =&gt; Array
        (
            [slug] =&gt; neurosymbolic
            [parent] =&gt; 
            [name] =&gt; neurosymbolic
            [canonical] =&gt; neurosymbolc ai
            [see] =&gt; 
            [description] =&gt; 
        )

    [newwave interface] =&gt; Array
        (
            [slug] =&gt; newwave interface
            [parent] =&gt; 
            [name] =&gt; newwave interface
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; NewWave was an alternative user interface for early versions of Windows on Hewlett-Packard hardware.  It included an {{intelligent agent}} portrayed as icon looking like a secret agent.
        )

    [newell, a] =&gt; Array
        (
            [slug] =&gt; newell, a
            [parent] =&gt; 
            [name] =&gt; newell, a
            [canonical] =&gt; newell, alan
            [see] =&gt; 
            [description] =&gt; 
        )

    [newell, alan] =&gt; Array
        (
            [slug] =&gt; newell, alan
            [parent] =&gt; 
            [name] =&gt; Newell, Alan
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; Alan Newell was one of the early pioneers of AI amonst other things proposing (with {{Herbert Simon}}) the {{physical symbol system hypothesis}} and developing (with {{John Laird}} and {{Paul Rosenbloom}}) the {{SOAR}} cognitive architecture.
        )

    [nlp] =&gt; Array
        (
            [slug] =&gt; nlp
            [parent] =&gt; 
            [name] =&gt; NLP
            [canonical] =&gt; 
            [see] =&gt; natural language processing
            [description] =&gt; 
        )

    [non-linear transformations] =&gt; Array
        (
            [slug] =&gt; non-linear transformations
            [parent] =&gt; 
            [name] =&gt; non-linear transformations
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; 
        )

    [non-locality] =&gt; Array
        (
            [slug] =&gt; non-locality
            [parent] =&gt; 
            [name] =&gt; non-locality
            [canonical] =&gt; 
            [see] =&gt; locality
            [description] =&gt; 
        )

    [non-monotonic reasoning] =&gt; Array
        (
            [slug] =&gt; non-monotonic reasoning
            [parent] =&gt; 
            [name] =&gt; non-monotonic reasoning
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; In non-monotonic reasoning new knowledge can change whether somethibg is infered to be true.  Non-montinic reasing often uses {{default rules}}.  For example, if Hoppy is a dog, and we know that dogs (usually) have four legs, then we might conclude that the statement &quot;Hoppy has four legs&quot; is true.  However, if we later learn that Hoppy had had a car accident as a puppy and had a leg amputated, we would then need to change our reasoning and instead conclude that &quot;Hoppy has four legs&quot; is false. 
        )

    [non-overlapping windows] =&gt; Array
        (
            [slug] =&gt; non-overlapping windows
            [parent] =&gt; 
            [name] =&gt; non-overlapping windows
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; 
        )

    [non-terminal symbols] =&gt; Array
        (
            [slug] =&gt; non-terminal symbols
            [parent] =&gt; 
            [name] =&gt; non-terminal symbols
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; In a {{grammar}}, a non-terminal symbol is one that may be expanded into further symbols, for example a noun-phrase may expand to a determiner and a noun.  In terms of a {{parse tree}}, non-terminal symbols usually label nodes wirthin the body of the tree, whereas {{terminal symbols}} label the leaves.
        )

    [non-zero-sum game] =&gt; Array
        (
            [slug] =&gt; non-zero-sum game
            [parent] =&gt; 
            [name] =&gt; non-zero-sum game
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; This is a game where one player's gain does not inevitably lead to the other player's loss.  This can be fully cooperative, for example, when several people need to solve parts of a larger puzzle, or more complicated, such as in the {{prisoners dilemma}}.  This is the opposite of a {{zero-sum game}}.
        )

    [non-zero-sum games] =&gt; Array
        (
            [slug] =&gt; non-zero-sum games
            [parent] =&gt; 
            [name] =&gt; non-zero-sum games
            [canonical] =&gt; non-zero-sum game
            [see] =&gt; 
            [description] =&gt; 
        )

    [nonlinear planning] =&gt; Array
        (
            [slug] =&gt; nonlinear planning
            [parent] =&gt; 
            [name] =&gt; nonlinear planning
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; 
        )

    [normal distribution] =&gt; Array
        (
            [slug] =&gt; normal distribution
            [parent] =&gt; 
            [name] =&gt; Normal distribution
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; The Normal or Gaussian distribution is the bell-shaped {{probability distribution}} that is often found in natural phenomena such as weights and heights.  Some forms of {{statistical anlaysis}} may assume that data is Normally distributed.  Also some algorithms may use the Normal distribution as part of theor processing, for example {{Gaussian filters}} for {{smoothing}} an image.  Normal distributions are very common because they arise when large numbers of nearly {{independent}} small items are added tgeher (e.g. the effect of large numbers of genes and environmental effects).  Even though they are common, it is important to remember that there are also many kinds of {{non-Normal distrubution}} for example {{power-law distributions}} that occur in much web and social network data.
        )

    [normalised information gain] =&gt; Array
        (
            [slug] =&gt; normalised information gain
            [parent] =&gt; 
            [name] =&gt; normalised information gain
            [canonical] =&gt; 
            [see] =&gt; information gain ratio
            [description] =&gt; 
        )

    [normalised mutual information] =&gt; Array
        (
            [slug] =&gt; normalised mutual information
            [parent] =&gt; 
            [name] =&gt; normalised mutual information
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; Normalised mutual information is an {{information measure}} used to assess the quality of {{clustering}} and {{classification algorithms}} ({{concept learning}}) in general.  If you split classified data into several groups, then the {{entropy}} of the result will be higher, even if the division is entirely {{random}} and unconnected to the target classification, simply becasue the reulting groups and hence probabilities are smaller.          It is obtained by dividing the raw {{information gain}} (gain(C,G)) by the sum of the entropy of the classification and decision grouping.  That is &lt;br&gt; &amp;nbsp; Normalised information gain &amp;nbsp; =  &amp;nbsp; gain(C,G)  /  ( H(C) + H(G) ) &lt;br&gt; where H(G) is the entropy of the division into the groups. &lt;br&gt;&lt;br&gt;. This is similar to but different from {{normalised information gain}}, as the latter just divides by H(G).  In practice this means normalised mutual information it is weighted slightly less against larger numbers of groups compared with  {{normalised information gain}}.
        )

    [normalization] =&gt; Array
        (
            [slug] =&gt; normalization
            [parent] =&gt; 
            [name] =&gt; normalization
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; 
        )

    [normally distributed] =&gt; Array
        (
            [slug] =&gt; normally distributed
            [parent] =&gt; 
            [name] =&gt; Normally distribution
            [canonical] =&gt; normal distribution
            [see] =&gt; 
            [description] =&gt; 
        )

    [noughts and crosses] =&gt; Array
        (
            [slug] =&gt; noughts and crosses
            [parent] =&gt; 
            [name] =&gt; noughts and crosses
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; The childs's game noughts and crosses (called tic-tac-toe in the USA) can be useful as. simple test case for computer {{game playing algorithms}}.  It is an example of a {{zero-sum game}}.
        )

    [obfuscation] =&gt; Array
        (
            [slug] =&gt; obfuscation
            [parent] =&gt; 
            [name] =&gt; obfuscation
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; Obfusication is when you try to conceal soemthing by burying it amongst other data or modifying it.  In particular {{privacy}} mechanisms may try to obfusicate perosnal details, for exmaple, slightly changing your birthday, or house number.
        )

    [object identification] =&gt; Array
        (
            [slug] =&gt; object identification
            [parent] =&gt; 
            [name] =&gt; object identification
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; 
            [subentries] =&gt; Array
                (
                    [object identification!bitmaps] =&gt; Array
                        (
                            [slug] =&gt; object identification!bitmaps
                            [parent] =&gt; object identification
                            [name] =&gt; bitmaps
                            [canonical] =&gt; 
                            [see] =&gt; 
                            [description] =&gt; 
                        )

                    [object identification!outlines] =&gt; Array
                        (
                            [slug] =&gt; object identification!outlines
                            [parent] =&gt; object identification
                            [name] =&gt; outlines
                            [canonical] =&gt; 
                            [see] =&gt; 
                            [description] =&gt; 
                        )

                    [object identification!paths] =&gt; Array
                        (
                            [slug] =&gt; object identification!paths
                            [parent] =&gt; object identification
                            [name] =&gt; paths
                            [canonical] =&gt; 
                            [see] =&gt; 
                            [description] =&gt; 
                        )

                    [object identification!summary statistics] =&gt; Array
                        (
                            [slug] =&gt; object identification!summary statistics
                            [parent] =&gt; object identification
                            [name] =&gt; summary statistics
                            [canonical] =&gt; 
                            [see] =&gt; 
                            [description] =&gt; 
                        )

                )

        )

    [object identification!bitmaps] =&gt; Array
        (
            [slug] =&gt; object identification!bitmaps
            [parent] =&gt; object identification
            [name] =&gt; bitmaps
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; 
        )

    [object identification!outlines] =&gt; Array
        (
            [slug] =&gt; object identification!outlines
            [parent] =&gt; object identification
            [name] =&gt; outlines
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; 
        )

    [object identification!paths] =&gt; Array
        (
            [slug] =&gt; object identification!paths
            [parent] =&gt; object identification
            [name] =&gt; paths
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; 
        )

    [object identification!summary statistics] =&gt; Array
        (
            [slug] =&gt; object identification!summary statistics
            [parent] =&gt; object identification
            [name] =&gt; summary statistics
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; 
        )

    [object recognition] =&gt; Array
        (
            [slug] =&gt; object recognition
            [parent] =&gt; 
            [name] =&gt; object recognition
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; 
        )

    [object-oriented architecture] =&gt; Array
        (
            [slug] =&gt; object-oriented architecture
            [parent] =&gt; 
            [name] =&gt; object-oriented architecture
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; 
        )

    [object-oriented programming] =&gt; Array
        (
            [slug] =&gt; object-oriented programming
            [parent] =&gt; 
            [name] =&gt; object-oriented programming
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; 
        )

    [objective function] =&gt; Array
        (
            [slug] =&gt; objective function
            [parent] =&gt; 
            [name] =&gt; objective function
            [canonical] =&gt; 
            [see] =&gt; fitness function
            [description] =&gt; 
        )

    [obstacle avoidance] =&gt; Array
        (
            [slug] =&gt; obstacle avoidance
            [parent] =&gt; 
            [name] =&gt; obstacle avoidance
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; 
        )

    [occam's razor] =&gt; Array
        (
            [slug] =&gt; occam's razor
            [parent] =&gt; 
            [name] =&gt; Occam's razor
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; Occam's razor (variously spelt Ockam, Ocham) is the principle that the simplest explanation is often the best or correct one.  It traces back to a 13th century monk William of Ockham, and is often stated as in Latin as &quot;pluralitas non est ponenda sine necessitate&quot;, which trabslates to &ldquo;plurality (or entities) should not be posited without necessity.&rdquo;  In AI this principle is often used to improve the {{generalisation}} after {{machine learning}} , for example {{pruning decision trees}} or the {{pinch points}} in {{deep neural networks}}.  It is closely related to issues of {{overfitting}} as models with too many free parameters often pick up irrelevant details of particular training examples.
        )

    [ocr] =&gt; Array
        (
            [slug] =&gt; ocr
            [parent] =&gt; 
            [name] =&gt; OCR (optical character recognition)
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; OCR (optical character recognition) is usually the first computer process to understand written or printed text.  Traditional methods involve matching against known patterns for letters, or first lookoing for features such as loops and lines. Neural networks are now used expecially for hand-written characters where ther is more variation.  A problem for both is matching chracters at different orientations, so some systems try to work out the general direction of text first and rotate the image, some rotate individual characters to match templates (called {{normalization}}).  Some neural networks are oreintation-indpeendent, by having layers which effectively perform this normalization.   Usuually the result of OCR is uncertain, both because of inherent ambiguities (is it zero or capital O, a small or large letter O, a 6 or a 9 upside down, VV or W?) and because the text may be poorly written or handwriting unusual.  This is resolved at higher levels of processing, for example when characters are grouped into word units they can be checked against a dictionary, allowing {{disambiguation}} of the lower level.  Typically some of this higher level processing is carried but within OCR resulting in plain text, but in some applications more raw confidence measuers are useful to allow yet higher-level dismbiguation, or highlighting of regions for hand checking.  
        )

    [octree] =&gt; Array
        (
            [slug] =&gt; octree
            [parent] =&gt; 
            [name] =&gt; octree
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; Octrees are the 3D version of {{quadtrees}}.  The octree is a tree-based data representation to store and index location information about a 3D image or area.  The root of the tree has eight branches corresponding to the eight {{octants}} of the image/area.  For example, if the area is a unit square, then the first branch is for the area with coprdinates (0&ndash;0.5,0&ndash;0.5,0&ndash;0.5), and the next one (0&ndash;0.5,0&ndash;0.5,0.5&ndash;1).  If there are no data points in any of these octants, then they are marked as empty.  Otherwise the points are allocated to the relevant octant and then the same process is used to build an octree of those.
        )

    [octrees] =&gt; Array
        (
            [slug] =&gt; octrees
            [parent] =&gt; 
            [name] =&gt; octrees
            [canonical] =&gt; octree
            [see] =&gt; 
            [description] =&gt; 
        )

    [oncue] =&gt; Array
        (
            [slug] =&gt; oncue
            [parent] =&gt; 
            [name] =&gt; oncue
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; 
        )

    [ontologies] =&gt; Array
        (
            [slug] =&gt; ontologies
            [parent] =&gt; 
            [name] =&gt; ontologies
            [canonical] =&gt; ontology
            [see] =&gt; 
            [description] =&gt; 
        )

    [ontology] =&gt; Array
        (
            [slug] =&gt; ontology
            [parent] =&gt; 
            [name] =&gt; ontology
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; 
        )

    [openai] =&gt; Array
        (
            [slug] =&gt; openai
            [parent] =&gt; 
            [name] =&gt; OpenAI
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; Open AI is the company behind the {{large language models}} {{GPT-3}} and {{GPT-4}} that power the {{chatbit}} {{chatGPT}}.  
        )

    [open ai] =&gt; Array
        (
            [slug] =&gt; open ai
            [parent] =&gt; 
            [name] =&gt; open AI
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; The term 'open AI' refers to AI technology and software that is open in the sense of {{open source}} or {{public domain}}.  This includes a variety of types of licence, and may refer to everything about a systme or jsut some aspects.  For example, one might make. a {{generative model}} available in the public domain, but not the training set that gave rise to it, nor the algorithms used to create it.  Not to be confused with the company {{OpenAI}}.
        )

    [open list] =&gt; Array
        (
            [slug] =&gt; open list
            [parent] =&gt; 
            [name] =&gt; open list
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; In {{search algorithms}} the open list is tothe (usually ordered) list of nodes that still need to be explored.  The basic {{breadth-first}} and {{depth-first}} search simply take nodes in turn from the front of the open list to explore next, they differ only in whether the children of nodes that are explored are added to the front or end of the open list.  More sophisticated algorithms (such as {{A*}}) may use more complex rules to decide which item to take next from the open list, often using {{heuristics} to guide the choice.  {{Graph search}} algorithms often also require a {{closed list}} recording the nodes that have already been considered.
        )

    [open world assumption] =&gt; Array
        (
            [slug] =&gt; open world assumption
            [parent] =&gt; 
            [name] =&gt; open world assumption
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; 
        )

    [open-loop control] =&gt; Array
        (
            [slug] =&gt; open-loop control
            [parent] =&gt; 
            [name] =&gt; open-loop control
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; 
        )

    [ops5] =&gt; Array
        (
            [slug] =&gt; ops5
            [parent] =&gt; 
            [name] =&gt; OPS5
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; OPS5 is an early {{production system}} language that was used in the {[expert system}} {{XCON}}.
        )

    [optical character recognition] =&gt; Array
        (
            [slug] =&gt; optical character recognition
            [parent] =&gt; 
            [name] =&gt; optical character recognition
            [canonical] =&gt; 
            [see] =&gt; ocr
            [description] =&gt; 
        )

    [optical flow] =&gt; Array
        (
            [slug] =&gt; optical flow
            [parent] =&gt; 
            [name] =&gt; optical flow
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; 
        )

    [optimal] =&gt; Array
        (
            [slug] =&gt; optimal
            [parent] =&gt; 
            [name] =&gt; optimal
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; 
        )

    [optimality] =&gt; Array
        (
            [slug] =&gt; optimality
            [parent] =&gt; 
            [name] =&gt; optimality
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; 
        )

    [optimization] =&gt; Array
        (
            [slug] =&gt; optimization
            [parent] =&gt; 
            [name] =&gt; optimization
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; 
        )

    [optional values] =&gt; Array
        (
            [slug] =&gt; optional values
            [parent] =&gt; 
            [name] =&gt; optional values
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; 
        )

    [outliers] =&gt; Array
        (
            [slug] =&gt; outliers
            [parent] =&gt; 
            [name] =&gt; outliers
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; Outliers are data items that are very unusual, for example having particularly large or small values of a numerical feature.. Outliers can be problematic, especially for statistical methods that use the {{arithmetic mean}}, {{variance}} or similar measures, all of which are quite sensitive to outliers.  This can also be a problem in pre-provessing for AI and ML algorithms if, for example, we use the mid-point of the full range of a feature as a way of reducing it a categorical large--small derived feature. One way to avoid these problems is to use metrics, such as the {{median}} or {{interquartile range}}, or to {{pre-whiten}} data such as scaling it to {{percentiles}}.  Another approach is to remove outliers form the data, for example, a common rule is to remove items where data values are more than twice the {{standard devoation}}.  It is important to have a fixed, pre-determined rule for this if possible to avoid {{cherry picking}} the removal of outliers that simpyl give results you don't like, and hence adding your own {{bias}} to the data.
        )

    [overfitting] =&gt; Array
        (
            [slug] =&gt; overfitting
            [parent] =&gt; 
            [name] =&gt; overfitting
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; Overfitting is when a {{machine learning}} or a {{statistical model}} starts to model very particular features of the {{training set}} rather than the more general features of the underlying populaton or phenomena the training set is intended to be representative of.  In the extreme the learning process precsiely mateches the training exmaples, but nothing else.  This can be intentional, as in the case of {{memoisation}}, but usually an accdient due to {{fitting}} too many {{model parameters}}.  In general there should be far fewer parameters (e.g. weights in a {{neural network}}) than features in the trainng data set.  For exmaple, if you have a thousand training items with ten features per item, then you should have far less than ten thousand (1000 items x 10 features per item) parameters.  By these metrics, some {{large language models}} appear to have too many parameters, and indeed do sometimes show features of overfitting such as directly reproducing passages of training text.  However, there are arguments that this is not so critical as early layers of {[deep neural networks}} are producing large amounts of {{non-linear diversity}}, which is then effectvely internally selected and pruned by later layers, especially at {{pinch points}}.
        )

    [overlearning] =&gt; Array
        (
            [slug] =&gt; overlearning
            [parent] =&gt; 
            [name] =&gt; overlearning
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; 
        )

    [ovid] =&gt; Array
        (
            [slug] =&gt; ovid
            [parent] =&gt; 
            [name] =&gt; Ovid
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; Ovid was a Roman poet, who, amongst other works, wrote about {{Pygmalion}} a sculptor who fell in love with a statue he had carved. The state was brought to life by Aphrodite at his request. The story is often used as a parallel for makers of {{artificial intelligence}} and {{robots}}, especially those espousing {{strong AI}}.
        )

    [owl] =&gt; Array
        (
            [slug] =&gt; owl
            [parent] =&gt; 
            [name] =&gt; OWL
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; OWL (Web Ontology Language) is a family of notations for expressing properties of an {{RDF}} {{ontology}}, which vary in expressive power.  The simplest, {{OWLlite}}, can express whether instances of a class should have particlar properties (say that animals have a number of legs) or that a class is a sub-class of another (e.g. Dalmation is a kind of dog).  More complex variants allow oen to declare, for example, that a class can only have a fixed set of instances, or to use levels of meta-modelling where classes can be treated as instances.  In general, simpler variants are easier to deal with and check computationally, but are less expressive.
        )

    [pagerank] =&gt; Array
        (
            [slug] =&gt; pagerank
            [parent] =&gt; 
            [name] =&gt; PageRank
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; 
        )

    [pagerank algorithm] =&gt; Array
        (
            [slug] =&gt; pagerank algorithm
            [parent] =&gt; 
            [name] =&gt; pagerank algorithm
            [canonical] =&gt; pagerank
            [see] =&gt; 
            [description] =&gt; 
        )

    [parallax] =&gt; Array
        (
            [slug] =&gt; parallax
            [parent] =&gt; 
            [name] =&gt; parallax
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; Parallax is the way that the relative position of objects can appear different from different points of observation.  In particular for machine (and human) vision a more distant object may appear to the left of a closer object when viewed from your left eye and to the right when seen from your right eye, this allowing {{stereo vision}}.  It also is the reason that distant objects seem to follow you as you move, a principle used in some forms of {{active vision}} systems.
        )

    [parallel processing] =&gt; Array
        (
            [slug] =&gt; parallel processing
            [parent] =&gt; 
            [name] =&gt; parallel processing
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; Parallel processing is when several computers or computing elemnts within a single computer can run computation at the same time (in parallel).  It may occur at a fairly large scale, such as tow separate computers or two cores in a single CPU.  It may also occur at a fine scale, for example the way a {{GPU}} does the same operation to many streams of data similtaneously.   Parallel processing is often divided into two main types {{SIMD}} (single instruction, multiple data) when the same things happens to lots of data in parallel, as in the case of the {{GPU}}; and {{MIMD}} (multiple instruction, multiple data) when different code may operate at teh same time, as in the case of CPU cores.
        )

    [parallelism] =&gt; Array
        (
            [slug] =&gt; parallelism
            [parent] =&gt; 
            [name] =&gt; parallelism
            [canonical] =&gt; parallel processing
            [see] =&gt; 
            [description] =&gt; 
        )

    [pareto frontier] =&gt; Array
        (
            [slug] =&gt; pareto frontier
            [parent] =&gt; 
            [name] =&gt; pareto frontier
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; The Pareto frontier is the set of all Pareto-optimal solutions.  For example, in the figure below, S1, S4, S5 and S6 are on the Pareto frontier.  The Pareto-frontier may be useful in {{visualising}} the trade-offs in a {{multi-objective optimisation}}
        )

    [pareto, vilfredo] =&gt; Array
        (
            [slug] =&gt; pareto, vilfredo
            [parent] =&gt; 
            [name] =&gt; Pareto, Vilfredo
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; 
        )

    [pareto-optimal] =&gt; Array
        (
            [slug] =&gt; pareto-optimal
            [parent] =&gt; 
            [name] =&gt; Pareto-optimal
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; In {{multi-objective optimisation}}, a value is said to be Pareto-optimal is there is no alternative value that is better for all criteria.  Typically there may be values where one is better on one criterion and the other on another.  For example, when considering ways to read, we may think paper books are best for ease of thumbing through pages rapidly, but that ebooks are best when searching.  There is no technology that is best on all criteria, but both paper books and ebooks are pareto-optimal, as neither is better than the other on all citeria.
        )

    [pareto-optimality] =&gt; Array
        (
            [slug] =&gt; pareto-optimality
            [parent] =&gt; 
            [name] =&gt; pareto-optimality
            [canonical] =&gt; pareto-optimal
            [see] =&gt; 
            [description] =&gt; 
        )

    [parse tree] =&gt; Array
        (
            [slug] =&gt; parse tree
            [parent] =&gt; 
            [name] =&gt; parse tree
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; A parse tree is often the output of a {[parser}}.  It typically follows the {{grammar}} of the underlying natural language or formal notation, with higher level nodes tagged with {{non-terminal symbols}} of the {{grammar}} and leaves tagged with {{terminal symbols}} and connected to actual words in the text.  In a formal gramer, such as for programming language a mathematical formulae, the parse tree may be  executes or evaluated dircetly by an interpreter or compiled into lower-level code or later execution.
        )

    [parser] =&gt; Array
        (
            [slug] =&gt; parser
            [parent] =&gt; 
            [name] =&gt; parser
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; The term 'parser' can be used in several, related, senses.  In {{natural language processing}} a parser is the code  that execute during {{syntactic analysis}} to match the {{gramamr}} of the language.  The term is also used in {{pattern matching}} systems, for example, in {{entity recognition}} used in {{data preparation}} or {{data detectors}} to match particular kinds of data such as serial numbers, addresses or formulae.  It can also be used in compilers and interpreters of programming languages when matching code to the formal grammar of the language.
        )

    [part-of-speech tagger] =&gt; Array
        (
            [slug] =&gt; part-of-speech tagger
            [parent] =&gt; 
            [name] =&gt; part-of-speech tagger
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; A part-of-speach tagger takes text and annotates it with tags representing parts of speach such as 'noun phrase' or 'adjective'.  It is often used as an early stage of {{natural language pricessing}} before {{syntactic analysis}} and further stages.
        )

    [partial state] =&gt; Array
        (
            [slug] =&gt; partial state
            [parent] =&gt; 
            [name] =&gt; partial state
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; 
        )

    [pattern-based recognition] =&gt; Array
        (
            [slug] =&gt; pattern-based recognition
            [parent] =&gt; 
            [name] =&gt; pattern-based recognition
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; 
        )

    [pen-based systems] =&gt; Array
        (
            [slug] =&gt; pen-based systems
            [parent] =&gt; 
            [name] =&gt; pen-based systems
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; Pen-based systems refers to devices that use a pen or stylus for input.  They usually require some form of {{handwriting recognition}} for text input and sometimes {{gesture recognition}} for actions such as deleting text.
        )

    [penrose] =&gt; Array
        (
            [slug] =&gt; penrose
            [parent] =&gt; 
            [name] =&gt; penrose
            [canonical] =&gt; penrose, roger
            [see] =&gt; 
            [description] =&gt; 
        )

    [penrose, roger] =&gt; Array
        (
            [slug] =&gt; penrose, roger
            [parent] =&gt; 
            [name] =&gt; Penrose, Roger
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; Roger Penrose is a physicist famous for his Nobel Prize winning work on black holes and also author on the mechanisms and philosophy of {{consciosness}}.
        )

    [penumbra] =&gt; Array
        (
            [slug] =&gt; penumbra
            [parent] =&gt; 
            [name] =&gt; penumbra
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; The penumbra is the area under part shadow, typically around the edge of the full shadow.  In sunlight these are points on the ground where part of the sun is visble.  Metaphorically, we can think of data items close to  {{decison boundary}} as being in the penumbra, as small changes to a few features might change the classification.
        )

    [percentiles] =&gt; Array
        (
            [slug] =&gt; percentiles
            [parent] =&gt; 
            [name] =&gt; percentiles
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; Percentiles are the pnts in a set of numerical values corresponding to how far (as a percentage) they sit through the ordered lost.  For example, the 10 percentile is the point such that 10% of the data values are below and 90% above.  There are special cases such as the {{median}} which is at the half way point, or 50 percentile, and the {{lower quartile}} and {{upper quartile}} that lie at the 25% and 75% poinrs respectively.  It can be useful to transform data into percentiles during {{data preparation}} as it effectively turns the data into a {{uniform distribution}} and reduces the undue influence of {{outliers}}.  It can also be useful for {{long-tail distributions}} as it reduces the impact of the small number of exterme tail values.
        )

    [perception] =&gt; Array
        (
            [slug] =&gt; perception
            [parent] =&gt; 
            [name] =&gt; perception
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; Perception refers to the process and experience by which the raw {{sensation}} of our senses (such as sight and sound) becomon meaningful.  AI tries to emulate these proceses in {{computer vision}} and {{natural languag eprocessing}} as low-level processing is sued to drive more {{semantic}} understanding.  While it is easiest to thnk of a bottom up process from raw {{sensation}} to perception, in fact in human cognituon there are top-down processes at work, we see what we expewct to see, which can aid disambiguation, but also can have problems if high-level expectations become locked in.  Computer perception also is most successful when there is a mix of bottom-up and top-down processes.
        )

    [perceptron] =&gt; Array
        (
            [slug] =&gt; perceptron
            [parent] =&gt; 
            [name] =&gt; perceptron
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; The perceptron is the earliest kind of {{artificial neuron}}.  It has a number of inputs and weights for each.  In operation the iputs are multipled by the weights and summed to gib and input {{caruvation}}.  This is then compared with a {{threshold function}}, and if the input activation exceeds the threshold, the perceptron fires. A simple percepttron arranged ina. single layer acnnot solve complex problems such as the {{XOR problem}}, but multi-layer perceptrons are hard to train.  For many years this limited practicla developemnt of {{arificial neural networks}} until the adoption of {{non-linear threshold functons}} and the development {{backpropogation}}.
        )

    [perceptrons] =&gt; Array
        (
            [slug] =&gt; perceptrons
            [parent] =&gt; 
            [name] =&gt; perceptrons
            [canonical] =&gt; perceptron
            [see] =&gt; 
            [description] =&gt; 
        )

    [pereira, f] =&gt; Array
        (
            [slug] =&gt; pereira, f
            [parent] =&gt; 
            [name] =&gt; Pereira, F.
            [canonical] =&gt; pereira, fernando
            [see] =&gt; 
            [description] =&gt; 
        )

    [pereira, fernando] =&gt; Array
        (
            [slug] =&gt; pereira, fernando
            [parent] =&gt; 
            [name] =&gt; Pereira, Fernando
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; Fernando Pereira is an early AI researcher, who amongst other things, was one of the creators of \term{definite clause grammar}, and is now VP and Engineering Fellow at Google.
        )

    [personal information management] =&gt; Array
        (
            [slug] =&gt; personal information management
            [parent] =&gt; 
            [name] =&gt; personal information management
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; Personal information management (PIM) is the study of the way people gather, organise, retreive and reuse their own information.  It is usually in contrast to organisational data, such as large databases or customer relations managment (CRM) software, as PIM is concerned more with small data, often of an informal or individual nature.  However, PIM software may be used by people in organisations, for example, contact lists, as well as in domestic settings, for example budget planners.  PIM has often been the target application for work on {{intelligent agents}}.
        )

    [personal knowledge management] =&gt; Array
        (
            [slug] =&gt; personal knowledge management
            [parent] =&gt; 
            [name] =&gt; personal knowledge management
            [canonical] =&gt; 
            [see] =&gt; personal information management
            [description] =&gt; 
        )

    [perturbation] =&gt; Array
        (
            [slug] =&gt; perturbation
            [parent] =&gt; 
            [name] =&gt; perturbation
            [canonical] =&gt; 
            [see] =&gt; perturbation techniques
            [description] =&gt; 
        )

    [perturbation techniques] =&gt; Array
        (
            [slug] =&gt; perturbation techniques
            [parent] =&gt; 
            [name] =&gt; perturbation techniques
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; Perturbation techniques make small changes to inputs to a model or algorithm and the. track how they change the output.  They are {{black-box methods}} in that they do not need to know anything about the internal workings of the model or algorithm.  Perturbatuon techniques can be used to measure the {[sensitivity}} of a model to changes and hence the {{uncertainty}} of the model.  They can also be really useful for {{explainable AI}}, particularly when applied to provide a ({local explanation}} for a specific decision.
        )

    [phase] =&gt; Array
        (
            [slug] =&gt; phase
            [parent] =&gt; 
            [name] =&gt; phase
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; In waves and frequency-based analaysis, such as {{Fourier transforms}}, phase refers to the time offset of the wave as opposed to the {{wavelength}} (how often it movs form higb to low and back) and the {{amplitude}} how high or low it goes.
        )

    [phenomenological] =&gt; Array
        (
            [slug] =&gt; phenomenological
            [parent] =&gt; 
            [name] =&gt; phenomenological
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; Phenomenological approaches take into account {{subjective experience}} as well as external behaviour.  For example, the experience of seeing red, not merely the 700 nanometre wavelength that excites the red-sensing cones in the human eye.
        )

    [phenotype] =&gt; Array
        (
            [slug] =&gt; phenotype
            [parent] =&gt; 
            [name] =&gt; phenotype
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; In biological {{genetics}} the pheonotype is the observable characteristics of the organism in contrast to the {{genotype}} which is the actual gene structure that encodes it.  In computational {{genetic algorithms}} there is a similar difference between the way in which the state is encoded, for example, the locations of appliances ina. kitchen, and the way that gives rise to, say, plumbing and wiring issues within the kitchen.  Often the mapping between {{genotype}} and pheonotype is effectively a one-way function, where it is hard to work out the geotype given the phenotype.  However, for a {{genetic algorithm}} to work it is essential that there is at least some pattern to the relationship otherwise it will be no better than a {{Monte Carlo algorithm}}.
        )

    [phone slave] =&gt; Array
        (
            [slug] =&gt; phone slave
            [parent] =&gt; 
            [name] =&gt; Phone Slave
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; Phone Slave was an early envisionment that used state of the art {{speech processing}} in the 1980s to show how an telephone answering machine could be transformed.  It prefigured many of the issues of {{speech-based interfaces}} such as how to avoid accidentally invoking commands when talking to other people in the environment.
        )

    [phonetic typewriter] =&gt; Array
        (
            [slug] =&gt; phonetic typewriter
            [parent] =&gt; 
            [name] =&gt; phonetic typewriter
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; The phonetic typewriter was an early example of the use of {{Kohonen nets}}.
        )

    [physical symbol system hypothesis] =&gt; Array
        (
            [slug] =&gt; physical symbol system hypothesis
            [parent] =&gt; 
            [name] =&gt; physical symbol system hypothesis
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; The physical symbol system hypothesis, forst stated by {{Alan Newell}} in 1976, states that &quot;a physical symbol system has the necessary and sufficient means for general intelligent
action&quot;.  A physical symbol system in this sense comprises a set of tokens or symbols with rules to manipulate them.  In computer terms we can think of digital representatiosm, but the term can also be applied to physical tokens, as envisaged in {{Searle's Chinese room}}.  It is related to notions of {{physicalism}} in the {{philosophy of intelligence}} and {{consciousness}}, but is sometimes given a tighter meaning assumes a higher-level symbol system as found in {{symbolic AI}} as opposed to {{subsymbolic}} {{neural networks}}; however the altter can be argued to also be manioulating 'symbols' on a broad sense of the term, just vast numbers of them.
        )

    [physicalism] =&gt; Array
        (
            [slug] =&gt; physicalism
            [parent] =&gt; 
            [name] =&gt; physicalism
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; Physicalism is the {{philosophical}} tenet that everything can be explained by physical processes.  This isthe domonant view in modern {{philosophy}} and {{physics}}.  This is closely associated with views that all intelligence and even consciousness can be explained by neurones in the brain acting as a form of {{information processing}} engine, and one of the arguments for {{strong AI}},   Physicalism is also losely related to the {{physical symbol system hypothesis}}, although the latter is often not taken to include {[sub-symbolic}} processing.    Within philosophy, physicalism and  {{materialism}} may be used interchangably, but the latter may sometimes be taken in a slightly more restricted sense of describing things interms of matter not including other aspects of the physical world such as energy fields.
        )

    [piecewise linear] =&gt; Array
        (
            [slug] =&gt; piecewise linear
            [parent] =&gt; 
            [name] =&gt; piecewise linear
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; A piecewise linear fit or model is one that uses a collection of linear patches (in 2D straight lines) to approximate a curved line or (hyper)surface.
        )

    [piecewise linear models] =&gt; Array
        (
            [slug] =&gt; piecewise linear models
            [parent] =&gt; 
            [name] =&gt; piecewise linear models
            [canonical] =&gt; piecewise linear
            [see] =&gt; 
            [description] =&gt; 
        )

    [pinch point] =&gt; Array
        (
            [slug] =&gt; pinch point
            [parent] =&gt; 
            [name] =&gt; pinch point
            [canonical] =&gt; pinch-point layer
            [see] =&gt; 
            [description] =&gt; 
        )

    [pinch points] =&gt; Array
        (
            [slug] =&gt; pinch points
            [parent] =&gt; 
            [name] =&gt; pinch points
            [canonical] =&gt; pinch-point layer
            [see] =&gt; 
            [description] =&gt; 
        )

    [pinch-point] =&gt; Array
        (
            [slug] =&gt; pinch-point
            [parent] =&gt; 
            [name] =&gt; pinch-point
            [canonical] =&gt; pinch-point layer
            [see] =&gt; 
            [description] =&gt; 
        )

    [pinch-point layer] =&gt; Array
        (
            [slug] =&gt; pinch-point layer
            [parent] =&gt; 
            [name] =&gt; pinch-point layer
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; Often a {{deep neural network}} includes one or more layers with far fewer nodes than others.  These pinch points often represent compact representations of {{features}}.  Thery may also be useful as part of techniques for {{explainable AI}}.
        )

    [placing dominoes] =&gt; Array
        (
            [slug] =&gt; placing dominoes
            [parent] =&gt; 
            [name] =&gt; placing dominoes
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; Placing domonoes is a simple (and rahter boring) game played on a squared board with {{dominoes}} to illustrate {{game trees}}.
        )

    [plateau] =&gt; Array
        (
            [slug] =&gt; plateau
            [parent] =&gt; 
            [name] =&gt; plateau
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; A pleateau in a {{fitness landscape}} is a region where there is little or no change on the {{fitness function}}.  This can a problem for methdos that use any form of {{hill climbing}} as small changes make no difference and hence it is not clear what changes to meke in order to improve the fitness.
        )

    [plateaux] =&gt; Array
        (
            [slug] =&gt; plateaux
            [parent] =&gt; 
            [name] =&gt; plateaux
            [canonical] =&gt; plateau
            [see] =&gt; 
            [description] =&gt; 
        )

    [pogoing robots] =&gt; Array
        (
            [slug] =&gt; pogoing robots
            [parent] =&gt; 
            [name] =&gt; Pogoing robots
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; Robots that hop on a single leg, like a child on a Pogo stick, are not very practically useful, but are a good test of control algorithms.  In particular, often {{fuzzy algorithms}} perform better than those trying work out the full dynamics.
        )

    [pole balancing] =&gt; Array
        (
            [slug] =&gt; pole balancing
            [parent] =&gt; 
            [name] =&gt; pole balancing
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; A simple control task is to move a trolly back and forth while balancing a pole on top.  This is not practically useful, but helps test different control strategies.  In particular, it has been used as a test case for {{reinforcement learning}}.
        )

    [policy network] =&gt; Array
        (
            [slug] =&gt; policy network
            [parent] =&gt; 
            [name] =&gt; policy network
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; 
        )

    [poorly constrained] =&gt; Array
        (
            [slug] =&gt; poorly constrained
            [parent] =&gt; 
            [name] =&gt; poorly constrained
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; 
        )

    [poplog] =&gt; Array
        (
            [slug] =&gt; poplog
            [parent] =&gt; 
            [name] =&gt; Poplog
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; Poplog is multi-paradigm language allowing both logic and impertative styles of programming in the same system.  It was developed in the early 1980s as an implementation of {{Prolog}} in POP-11, and is still under ongoing development.
        )

    [position independent] =&gt; Array
        (
            [slug] =&gt; position independent
            [parent] =&gt; 
            [name] =&gt; position independent
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; In {{computer vision}} and {{image processing}}, we say a method to recognise an object in an image is position independent if it works no matter where the target is within the image.  This may be achieved by training on lots of different offsets of the same images, by centering images on pertinent features, or by using models that are intrinsically position independent.  In additoon we often seek {{scale independence}} and {{rotational independence}} (it doesnt matter how big or what anfgle the object is). 
        )

    [power law distribution] =&gt; Array
        (
            [slug] =&gt; power law distribution
            [parent] =&gt; 
            [name] =&gt; power law distribution
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; 
        )

    [pragmatic analysis] =&gt; Array
        (
            [slug] =&gt; pragmatic analysis
            [parent] =&gt; 
            [name] =&gt; pragmatic analysis
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; Pragmatic analysis is the final stage in {{natural language processing}} which tries to make sense of an utterance in {{context}}.  This may in part be used to disambiguate an utterance, for example, the word 'America' may mena the whole continant, North America as a whole or the USA depending on the speaker and context.   This can also be important for understanding what an utterance is &lt;em&gt;for&lt;/em&gt;.  For example, the statement &quot;it is cold&quot; can be interpretted as a simple fact about the temperature in the room, or as a {{speech act}} that means &quot;please shut the door&quot;.   Pragmatic analysis may follow {{semantic analysis}} in a {{natural language processing}} process, but also may be addeessed directly, for example by looking for particular patterns or terms in text.
        )

    [pre-processing] =&gt; Array
        (
            [slug] =&gt; pre-processing
            [parent] =&gt; 
            [name] =&gt; pre-processing
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; 
        )

    [pre-processor] =&gt; Array
        (
            [slug] =&gt; pre-processor
            [parent] =&gt; 
            [name] =&gt; pre-processor
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; 
        )

    [pre-whiten] =&gt; Array
        (
            [slug] =&gt; pre-whiten
            [parent] =&gt; 
            [name] =&gt; pre-whiten
            [canonical] =&gt; pre-whitening
            [see] =&gt; 
            [description] =&gt; 
        )

    [pre-whitening] =&gt; Array
        (
            [slug] =&gt; pre-whitening
            [parent] =&gt; 
            [name] =&gt; pre-whitening
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; Often data values are transformed to make them more like a standard distribution. This process is called pre-whetning, especially for temporal data when the goal is to transform the data to be more like {{white noise}}.  Other examples are transforming data to make it more like a {{normal distribution}} of {{log transforms}} of timing data.
        )

    [precision] =&gt; Array
        (
            [slug] =&gt; precision
            [parent] =&gt; 
            [name] =&gt; precision
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; When making a binary classification or decision, such as a disease diagnosis, precision is the {{probability}} that if we make a positive decision, it turns out out to be correct, for example, if we diagnose a person with the disease they really have it.  If a decision has a high precision, then the probability of a {{false positive}} is low.  Often increasing precision reduces {{recall}}; this is called the {{precision-recall trade-off}}.
        )

    [precision--recall trade-off] =&gt; Array
        (
            [slug] =&gt; precision--recall trade-off
            [parent] =&gt; 
            [name] =&gt; precision--recall trade-off
            [canonical] =&gt; precision-recall trade-off
            [see] =&gt; 
            [description] =&gt; 
        )

    [precision-recall trade-off] =&gt; Array
        (
            [slug] =&gt; precision-recall trade-off
            [parent] =&gt; 
            [name] =&gt; precision&amp;ndash;recall trade-off
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; Often algorithms give some sort of confidence measure for making a desicison, for example, a cancer diagnois.   We can then make the decision based on a threshold value, making a postive diagnosis if the confidence exceeds the threshold..  If the threshold of confidence is very high, we ensure high {{precision}}, that is if we make diagnosis, then it is almost certainyl right; but this then means that we are likley to miss some potential cases ({{false positives}}), that is low {{recall}}.  Alternateively a low threshold value will give high {[recall}}, we are unlikely to miss positive cases, but more likely to falsely diagnose someone ({{false negative}}), that is low {{precision}}.  In general, there is a trade-off between {{precision}} and {{recall}} and the correct balance is a {{cost--benefit}} trade-off depending on the relative impact of {{false positives}} and {{false negatives}}.
        )

    [predicate calculus] =&gt; Array
        (
            [slug] =&gt; predicate calculus
            [parent] =&gt; 
            [name] =&gt; predicate calculus
            [canonical] =&gt; predicate logic
            [see] =&gt; 
            [description] =&gt; 
        )

    [predicate logic] =&gt; Array
        (
            [slug] =&gt; predicate logic
            [parent] =&gt; 
            [name] =&gt; predicate logic
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; Predicate logic (also known as first-order logic or precdiate caculus) allows logical formulae including variables (such as X, Y, Person, Parent)  and predicates (such as is_child_of(Person, Parent).  It does not allow higher-order formulae, that is predicte symbols cannot also be used as variables.
        )

    [principal components] =&gt; Array
        (
            [slug] =&gt; principal components
            [parent] =&gt; 
            [name] =&gt; principal components
            [canonical] =&gt; principal components analysis
            [see] =&gt; 
            [description] =&gt; 
        )

    [principal components analysis] =&gt; Array
        (
            [slug] =&gt; principal components analysis
            [parent] =&gt; 
            [name] =&gt; principal components analysis
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; Principal components analysis attempts to find a smaller dimensional subspace that captures the majority of the varation in a dataset.  It is a {{statistical technique}} that us used in various ways in AI.  Witin {{explainable AI}} it may be used to help interpret complex data, such as internal layers of a {[deep neural network}}.  It can also be used as a {{dimensional reduction}} technique during {{data preparation}}.  One way to calculate the principal components of a data set is first to calcualte the feature cross-correltion matrix, that is the {{matrix}} comprising of the {{correlations}} between different features.  The principal components are then the N {{eigenvectors}} with the largest {{eignevalues}}.   In. particular, the first principal component is the {{principal eigenvector}}. One can decide in advance how many principal components are required, or inspect them one by one choosing a point to stop where the eigenvalues drop below a chosen level. 
        )

    [principal eigenvector] =&gt; Array
        (
            [slug] =&gt; principal eigenvector
            [parent] =&gt; 
            [name] =&gt; principal eigenvector
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; The principal eigenvector is the {{eigenvector}} with the largest {[eigenvalue}}.  It is the direction that has the greatest variation within the datset.
        )

    [prisoners' dilemma] =&gt; Array
        (
            [slug] =&gt; prisoners' dilemma
            [parent] =&gt; 
            [name] =&gt; prisoners' dilemma
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; The prisoner's dilemma is a classic {{game theory}} problem.  Two prisoners are being separately interviewed by the police.  If one confeses and the other does not they one who confesses gets a reduced sentance of 1 year and the other gets a very long sentence of 10 years, if both confess they both get 5 years, but if neither confess they are freed.  The most risk-free choice for each prisoner is to confess (worst case 5 years), but the best thing is for neither to confess.  This is an example of a {{non-zero-sum game}} as the rewards/penalties are not a strict trade-off between the parties. It also shows the importance of communication.
        )

    [privacy] =&gt; Array
        (
            [slug] =&gt; privacy
            [parent] =&gt; 
            [name] =&gt; privacy
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; Privacy is about the ability or right of people to control what information or physical knowledge about themselves is available to others.  More broadly it is associated with {{personal integrity}} &ndash;&ndash; being in control of ones own body, mind and social relationships.  Many countries have {[data protection}} legislation which restricts those collecting or holding data.  Digital systems create threwats to privacy both through the accidental leakage of information or through deliberate or ill-considered sharing by others, such as personal photographs.  The harms can be financial (e.g. {{identity theft}}), social (e.g. embarassment), and psychological (e.g. feeling exposed).  Often provacy is seen purely as a form of {{secrecy}} of personal information, but the invasion of {{personal integrity}} is broader: for example {{deep fakes}} that create nude or other forms of false appearance or the use of aggregated data for purposes that the data subject would not have assented to.
        )

    [privacy preserving algorithms] =&gt; Array
        (
            [slug] =&gt; privacy preserving algorithms
            [parent] =&gt; 
            [name] =&gt; privacy preserving algorithms
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; Privacy preserving algorithms attempt to perform machine learning on personal data without transfering the raw information to a central location, usually by processing locally on a user's own machine.  For example, deep neural networks make small differnce the the internal weights based on each training example.  Rather than send all of the training examples to a central location for processing, each person's local computer can work out the changes in the weights (known as a {{delta}}) and send these deltas where they can be combined to give overall learning.  It is hard to ensure total {{privacy}} as in some circumstances it has been shown that the original data can be reccovered from deltas.
        )

    [probabilistic approaches] =&gt; Array
        (
            [slug] =&gt; probabilistic approaches
            [parent] =&gt; 
            [name] =&gt; probabilistic approaches
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; 
        )

    [probabilistic process] =&gt; Array
        (
            [slug] =&gt; probabilistic process
            [parent] =&gt; 
            [name] =&gt; probabilistic process
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; 
        )

    [probabilistic reasoning] =&gt; Array
        (
            [slug] =&gt; probabilistic reasoning
            [parent] =&gt; 
            [name] =&gt; probabilistic reasoning
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; 
        )

    [probability] =&gt; Array
        (
            [slug] =&gt; probability
            [parent] =&gt; 
            [name] =&gt; probability
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; 
        )

    [probability density] =&gt; Array
        (
            [slug] =&gt; probability density
            [parent] =&gt; 
            [name] =&gt; probability density
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; 
        )

    [procedural knowledge] =&gt; Array
        (
            [slug] =&gt; procedural knowledge
            [parent] =&gt; 
            [name] =&gt; procedural knowledge
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; Procedural knowledge encodes some sort of process or algorithm to create an answer in contrast to {{decarative knowledge}}.  This distinction applies to {{human reasing}}, for example you might know how to add up large numbers on an abacus (procedural knowledge) wheras another person might simply remember that 12+23 is 35 ({{decarative knowledge}}).  It also applies to {{automated reasoning}} , for example the use of {{grammars}} and {{parsing}} in {{natural language processing}} (procedural knowledge) vs. looking things up in large databases ({{decarative knowledge}}). 
        )

    [proceduralisation] =&gt; Array
        (
            [slug] =&gt; proceduralisation
            [parent] =&gt; 
            [name] =&gt; proceduralisation
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; Proceduralisation is the process through which animals and also machine learning takes common sequences of actions, whether physical or mental, and combines them into a fixed unit, which is then one of the building blocks for future learning and action.  For example, a human guitar player may spend some time learning anew finger sequence, but after a while it no longer feels like separate movements, but is a single practiced motif.  Similarly, an automated proof algorithm might repeatedly apply the same sequence of algebraic manipulations during proofs, and after a while combine this into a new proof operator.   The term also has a specific meaning in the {{ACT-R}} cognitive architecture.
        )

    [proceduralization] =&gt; Array
        (
            [slug] =&gt; proceduralization
            [parent] =&gt; 
            [name] =&gt; proceduralization
            [canonical] =&gt; proceduralisation
            [see] =&gt; 
            [description] =&gt; 
        )

    [production rules] =&gt; Array
        (
            [slug] =&gt; production rules
            [parent] =&gt; 
            [name] =&gt; production rules
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; A production rule is an IF &amp;lt;condition&amp;gt; THEN &amp;lt;action&amp;gt; rule as found in a {{production system}}.
        )

    [production system] =&gt; Array
        (
            [slug] =&gt; production system
            [parent] =&gt; 
            [name] =&gt; production system
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; A production system consists of three elements: (1). a database or working memory of facts, which can include temporary states such as &quot;Goal(open door)&quot;; (2) a set of ules of the form IF &amp;lt;condition&amp;gt; THEN &amp;lt;action&amp;gt;, where conditions are logical expressions about the working memory, and actions may include changes to the working memory such as &quot;ADD Goal(find key)&quot;; and (3) an interpretter, which matches the conditoons against the working memory and then executes the action for a matching rule.
            [subentries] =&gt; Array
                (
                    [production system!loans example] =&gt; Array
                        (
                            [slug] =&gt; production system!loans example
                            [parent] =&gt; production system
                            [name] =&gt; loans example
                            [canonical] =&gt; 
                            [see] =&gt; loans example
                            [description] =&gt; 
                        )

                )

        )

    [production system!database] =&gt; Array
        (
            [slug] =&gt; production system!database
            [parent] =&gt; production system
            [name] =&gt; database
            [canonical] =&gt; production system
            [see] =&gt; 
            [description] =&gt; 
        )

    [production system!interpreter] =&gt; Array
        (
            [slug] =&gt; production system!interpreter
            [parent] =&gt; production system
            [name] =&gt; interpreter
            [canonical] =&gt; production system
            [see] =&gt; 
            [description] =&gt; 
        )

    [production system!language] =&gt; Array
        (
            [slug] =&gt; production system!language
            [parent] =&gt; production system
            [name] =&gt; language
            [canonical] =&gt; production system
            [see] =&gt; 
            [description] =&gt; 
        )

    [production system!loans example] =&gt; Array
        (
            [slug] =&gt; production system!loans example
            [parent] =&gt; production system
            [name] =&gt; loans example
            [canonical] =&gt; 
            [see] =&gt; loans example
            [description] =&gt; 
        )

    [production system!rules] =&gt; Array
        (
            [slug] =&gt; production system!rules
            [parent] =&gt; production system
            [name] =&gt; rules
            [canonical] =&gt; production rules
            [see] =&gt; 
            [description] =&gt; 
        )

    [production system!working memory] =&gt; Array
        (
            [slug] =&gt; production system!working memory
            [parent] =&gt; production system
            [name] =&gt; working memory
            [canonical] =&gt; production system
            [see] =&gt; 
            [description] =&gt; 
        )

    [programmable user models] =&gt; Array
        (
            [slug] =&gt; programmable user models
            [parent] =&gt; 
            [name] =&gt; programmable user models
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; Programmable user models ({PUMS}} was a class of {{cognitive models}} based on the {{SOAR}} cognitive architecture.  The model was used to act as a surrogate user in testing of {{user interfaces}} or other user tasks such as arithmatic.  The purpose of PUMS was to emulate the user clsoely enough to be able to test deisgns of {{user interfaces}}, not to do the tasks to replace the user.
        )

    [programming-by-demonstration] =&gt; Array
        (
            [slug] =&gt; programming-by-demonstration
            [parent] =&gt; 
            [name] =&gt; programming-by-demonstration
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; While the term 'programming by demonstration' sounds as though it os about coding, in fac it was terned to desceribe systems that learn users tasks, notably the early {{intlligent agent}} {{EAGER}}.  A programming by demonstration system observes users actions and looks for patterns, repetations, etc. and then proposes ways to automate the task (the 'progamming' part of the name).
        )

    [prolog] =&gt; Array
        (
            [slug] =&gt; prolog
            [parent] =&gt; 
            [name] =&gt; Prolog
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; Prolog is a prigramming language that is based on first order {{predicate logic}}.  It uses Horn clause logic (predicates can be defined in terms of other predicates, but not arbitrary equations) and {{unification}} of variables, with {{backtracking}} when it reaches an impasse or tries to seek additional solutions.  The outcome of a Prolog query is a binding of variables or 'No' meaning no (further) solution can be found.  Prolog allows negation and has a {{closed world}} model of knowledge -- if a fact is not found it is assumed to be false.
        )

    [property inheritance] =&gt; Array
        (
            [slug] =&gt; property inheritance
            [parent] =&gt; 
            [name] =&gt; property inheritance
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; Property inheritance is when a property of a class, such as &quot;dogs bark&quot; is assumed to be also true of sub-classes (such as Beagle) and instances (such as Snoopy).  This means that, in the absence of evidence to the contrary, we can infer that Beagles bark and that Snoopy barks.  
        )

    [prospector] =&gt; Array
        (
            [slug] =&gt; prospector
            [parent] =&gt; 
            [name] =&gt; PROSPECTOR
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; PROSPECTOR was an expert system developed at Stanford in the
late 1970s  to evaluate geological sites for potential mineral deposits.  In 1984 it was
instrumental in discovering a molybdenum deposit worth 100 million dollars. &lt;br&gt;&lt;br&gt; PROSPECTOR  was based on a {{semantic network}} and used a combination of mostly {{forward chaining}} and some {{backward chaining}}.  It adopted a form of {{depth first search}} guided by the probabilities of different hypotheses and used {{Bayesian reasoning}}  to deal with uncertainty.  The {{dialogue component}} and {{explanation component}} used {{mixed control}}.
        )

    [provenance] =&gt; Array
        (
            [slug] =&gt; provenance
            [parent] =&gt; 
            [name] =&gt; provenance
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; Provenace is about the source of an object or information or the processes that created it.  For example, in the phsyical world, the provenance of an piece of antique furniture would include who designed and made it, who originally owned it, were it was bought and sold since.  Computationally, we may want to store this kind of information about a data item as part of its {{meta-information}}, and there are {{semantic web}} standards for storing provenance in {{RDF}}.  If the data item is refering to a phsyical thing (suc as antique furniture), then this may be exactly the kind of informationabove.  For other kinds of data it may include the source of the data, how it was collected or what datsource it came from; this can help assess the {{reliability}} of the data.  Outside the {{semantic web}}, during {{data preparation}} it is usually good to keep track of the processing steps that gave rise a to a particlar data file -- it is easy to forget!  Provevance can also be important in {{explainable AI}}, in order to increase the {{transparency}}  ofand hence {{trust}} in AI processes.
        )

    [proxy indicator] =&gt; Array
        (
            [slug] =&gt; proxy indicator
            [parent] =&gt; 
            [name] =&gt; proxy indicator
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; When considering potemtial {{bias}}, a proxy indicator is some combination of features that reveal the value of a {{protected characteristic}} such as gender or ethnic background.  For example, Days taken as holiday from work might be indicative of a person's religion, or subjects tkane at school may be related to gender.  Often it is assumed that if {{protected characteristics}} are removed form a dataset before it is used for training, then the reuslying {{machine learning}} system will be free of {{bias}}.  However, proxy indicators mean that decisions are still effectievly made based on the {{protected characteristic}}.
        )

    [pruning] =&gt; Array
        (
            [slug] =&gt; pruning
            [parent] =&gt; 
            [name] =&gt; pruning
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; Pruning is when one removes or ignores a part of a data structure. or process.  The term derives from horticultural pruning of dead or unwanted branches from a growing tree or bush.  Many of AI examples are about trees, on the computational sense, hence the use in AI.  One example of removal is to reduce the size of {{decison trees}} after and {{inductive learming}} algorithm such as {{ID3}}; pruning can remve branches that have low additional dscrininatiin and hence (typically) improve the {{generalisation}} of the tree.  An ignoring example is during {{A* search}} where the {{heuristic}} means one knows that a particular node cannot be on the best path, and hence can prune the node from the search of the graph.
            [subentries] =&gt; Array
                (
                    [pruning!alpha-beta] =&gt; Array
                        (
                            [slug] =&gt; pruning!alpha-beta
                            [parent] =&gt; pruning
                            [name] =&gt; alpha&amp;ndash;beta
                            [canonical] =&gt; 
                            [see] =&gt; alpha-beta pruning
                            [description] =&gt; 
                        )

                )

        )

    [pruning!alpha-beta] =&gt; Array
        (
            [slug] =&gt; pruning!alpha-beta
            [parent] =&gt; pruning
            [name] =&gt; alpha&amp;ndash;beta
            [canonical] =&gt; 
            [see] =&gt; alpha-beta pruning
            [description] =&gt; 
        )

    [pseudonymisation] =&gt; Array
        (
            [slug] =&gt; pseudonymisation
            [parent] =&gt; 
            [name] =&gt; pseudonymisation
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; When data is  {{pseudonymised}}, each person's name (or other personally identifiable labels) is replaced with a unique {{anonymous identifier}} .  This may be used when full {{anonymisation}} would ber too string.  For exampel, we may want to be able to reconnect processed data with an individual, perhaps to warn them of a disease.  It is also important when different datasets need to be linked together, we need to be able to tell which records refer to the same individual even if we do not know &lt;em&gt;which&lt;/em&gt; actual person is refered to.
        )

    [pseudonymised] =&gt; Array
        (
            [slug] =&gt; pseudonymised
            [parent] =&gt; 
            [name] =&gt; pseudonymised
            [canonical] =&gt; pseudonymisation
            [see] =&gt; 
            [description] =&gt; 
        )

    [pseudonomised data] =&gt; Array
        (
            [slug] =&gt; pseudonomised data
            [parent] =&gt; 
            [name] =&gt; pseudonomised data
            [canonical] =&gt; pseudonymisation
            [see] =&gt; 
            [description] =&gt; 
        )

    [pums] =&gt; Array
        (
            [slug] =&gt; pums
            [parent] =&gt; 
            [name] =&gt; PUMS
            [canonical] =&gt; programmable user models
            [see] =&gt; 
            [description] =&gt; 
        )

    [pygmalion] =&gt; Array
        (
            [slug] =&gt; pygmalion
            [parent] =&gt; 
            [name] =&gt; Pygmalion
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; Pygmalion was a sculptor in the poem 'Metamorphoses' by the Roman poet {{Ovid}}. Pygmalion fell in love with a statue he had carved and asked Aphrodite to bring the statue to life.  However, once a flesh and blood women.  The story is often used as a parallel for makers of {{artificial intelligence}} and {{robots}}, esepcially those espousing {[strong AI}}.
        )

    [python] =&gt; Array
        (
            [slug] =&gt; python
            [parent] =&gt; 
            [name] =&gt; Python
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; Python is a programming language that is heavily used in {{machine learning}}, in part because of the ectebsive libarraies that have been developed and in part because of the success of {{Jupyter notebook}}, which makes it easy to experiment, modify and share not just programs or the output of a session of actvity, but also the details of the process itself in a manner that can be replayed and modified by others.
        )

    [qbb] =&gt; Array
        (
            [slug] =&gt; qbb
            [parent] =&gt; 
            [name] =&gt; QBB
            [canonical] =&gt; query-by-browsing
            [see] =&gt; 
            [description] =&gt; 
        )

    [quadtree] =&gt; Array
        (
            [slug] =&gt; quadtree
            [parent] =&gt; 
            [name] =&gt; quadtree
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; The quadtree is a tree-based data representation to store and index location information about a 2D image or area.  The root of the tree has four branches corresponding to the four quadrants of the image.  If there are no data points in any of these quadrants, then they are marked as empty.  Otherwise the points are allocated to the relevant quadrant and then the same process is used to build a quadtree of those.
        )

    [quadtrees] =&gt; Array
        (
            [slug] =&gt; quadtrees
            [parent] =&gt; 
            [name] =&gt; quadtrees
            [canonical] =&gt; quadtree
            [see] =&gt; 
            [description] =&gt; 
        )

    [quadstores] =&gt; Array
        (
            [slug] =&gt; quadstores
            [parent] =&gt; 
            [name] =&gt; quadstores
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; Quadstores are a form of {{database}} used to store {{semantic web}} data.. {{RDF}} is encoded as {{subject-predicate-object triples}} and often stored in a {{triplestore}}.  A quadstore allows several graphs of RDF data to be stored in teh same database.  Each item in the quadstores is of the form ( graph URI, subject, predicate, object ) and it is possible to query over a particular graph, all graphs or other combinations of graphs.  The graph URI can itself be used as the subject in assertions allowing {{provenance}} to be represented, for example, the origin, authority or relaibility of  a particualr sub-collection (graph) of triples.
        )

    [qualia] =&gt; Array
        (
            [slug] =&gt; qualia
            [parent] =&gt; 
            [name] =&gt; qualia
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; 
        )

    [quantifiers] =&gt; Array
        (
            [slug] =&gt; quantifiers
            [parent] =&gt; 
            [name] =&gt; quantifiers
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; 
        )

    [quantum computing] =&gt; Array
        (
            [slug] =&gt; quantum computing
            [parent] =&gt; 
            [name] =&gt; quantum computing
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; Quantum computers use various forms of quantum-level processs as an explicit part of their hardware.   They make extensive use of {[qubits}} physcial elements that are in multiple superimposed states, sort of 0 and 1 at the same time.  Quantum effects mean that effectively the same pieces of phsyical hardware simulrtaneosly perofrm multiple veesions of the same calculations, in contrats to classicla {{parallel processing}} where several physically dstinct processors do work on parts or variants of an algorith at the same time.  Ths may be used ot modle quantum-level effects in phsyics, but can also be used in other kinds of algoriths.  In particular many of the techniques used in {{cryptography}} will become insecure as a quantum computer can effectively try billions of options of a key similtaneously.  In AI, one can imagie that certain kinds of {{neural network}} could benefit and one of the technologies for {{reservoir computing}}  is quantum; however practical applications are probably still some years off  Note that conventional computers may experience quantum effects such as tunneling between adjacent tracks, but this is usually regarded as a problem (indeed is one of the limits of convetional electronics), whereas quantum computers make deliberate use of these kinds of effect.
        )

    [quartile] =&gt; Array
        (
            [slug] =&gt; quartile
            [parent] =&gt; 
            [name] =&gt; quartile
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; The quartiles of a {{probability distribution}} refer to the points (or soemtimes the intervals between them) that divide the paramter into four pieces where the probability of each is exactly a quarter.  This means that on average 25% of the time the value will lie below the {{lower quartile}} and 25% of the time larger than the {{upper quartile}}.  Arguably the {{median}} can be regarded as the middle quartile, but more often the term is sued only for the lowert and upper quartile.  
        )

    [quasi-periodic] =&gt; Array
        (
            [slug] =&gt; quasi-periodic
            [parent] =&gt; 
            [name] =&gt; quasi-periodic
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; A {{quasi-periodic}} phenomenon is one that almost has a periodic behaviour, but is not entirely predictable.  They may occur when otherwise periodic processes have noise.  For example, imagine a pedulum clock in a very windy location, the pendulum naturally swings once a secomd, but the gusts and breezes mean it may drift, with some swings a little faster and some a little slower.  Critically, if two such pendula were in different environments and started at the same moment, after a while the pendula would both still be approximately once a second, but may be out of {{phase}} with one another. 
        )

    [query-by-browsing] =&gt; Array
        (
            [slug] =&gt; query-by-browsing
            [parent] =&gt; 
            [name] =&gt; Query-by-Browsing
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; Query-by-Browsing (QbB) is an intelligent database query interface that allows users to choose records of interst from a listing and then infers a query, for example in {{SQL}}.  This is of course the opposite of most database query interfaces (including so called 'query by example') where the user writes.a query as input and then sees the results as output.
        )

    [quinlan] =&gt; Array
        (
            [slug] =&gt; quinlan
            [parent] =&gt; 
            [name] =&gt; quinlan
            [canonical] =&gt; quinlan, ross
            [see] =&gt; 
            [description] =&gt; 
        )

    [quinlan, ross] =&gt; Array
        (
            [slug] =&gt; quinlan, ross
            [parent] =&gt; 
            [name] =&gt; Quinlan, Ross
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; Ross Quinlan developed the influential {{ID3}} and {{C4.5}} algorithms for inductive learning of {{decision trees}}
        )

    [radial basis functions] =&gt; Array
        (
            [slug] =&gt; radial basis functions
            [parent] =&gt; 
            [name] =&gt; radial basis functions
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; 
        )

    [random decision forest] =&gt; Array
        (
            [slug] =&gt; random decision forest
            [parent] =&gt; 
            [name] =&gt; random decision forest
            [canonical] =&gt; 
            [see] =&gt; random forest
            [description] =&gt; 
        )

    [random decision forests] =&gt; Array
        (
            [slug] =&gt; random decision forests
            [parent] =&gt; 
            [name] =&gt; random decision forests
            [canonical] =&gt; random decision forest
            [see] =&gt; 
            [description] =&gt; 
        )

    [random forest] =&gt; Array
        (
            [slug] =&gt; random forest
            [parent] =&gt; 
            [name] =&gt; random forest
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; A random forest is a form of {{ensemble method}} for machine learning.  Large numbers of {{decision trees}} are generated by choosing random subsets of features from random subsets of the training data.  The indivdiual trees each give their own results and a separate layer of processing combines the results from all of the trees to give a final result.  While compartively simple, random forests have proved highly effective in a wide variey of application domains.
        )

    [random forests] =&gt; Array
        (
            [slug] =&gt; random forests
            [parent] =&gt; 
            [name] =&gt; random forests
            [canonical] =&gt; random forest
            [see] =&gt; 
            [description] =&gt; 
        )

    [random walk] =&gt; Array
        (
            [slug] =&gt; random walk
            [parent] =&gt; 
            [name] =&gt; random walk
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; 
        )

    [randomising the order] =&gt; Array
        (
            [slug] =&gt; randomising the order
            [parent] =&gt; 
            [name] =&gt; randomising the order
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; 
        )

    [rbm] =&gt; Array
        (
            [slug] =&gt; rbm
            [parent] =&gt; 
            [name] =&gt; RBM
            [canonical] =&gt; 
            [see] =&gt; restricted boltzmann machine
            [description] =&gt; 
        )

    [rdf] =&gt; Array
        (
            [slug] =&gt; rdf
            [parent] =&gt; 
            [name] =&gt; RDF
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; RDF is the {{semantic web}} technology used to experess properties of web pages and in general {{URI}}s oncluding those representing things outside of the web.  RDF is expressed using {{subject-predicate-object}} tripples, where both the subject and predicate have to be {{URI}}s, but the object may also be soem form of literal value such as a number or string.  This allows one to express statements such as: &lt;br&gt; &amp;lt;https://alandix.com/academic/&amp;gt; &amp;lt;ex:is_about&amp;gt; &quot;Alan Dix&quot;
        )

    [rdfs] =&gt; Array
        (
            [slug] =&gt; rdfs
            [parent] =&gt; 
            [name] =&gt; RDFS
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; 
        )

    [reactive movement] =&gt; Array
        (
            [slug] =&gt; reactive movement
            [parent] =&gt; 
            [name] =&gt; reactive movement
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; 
        )

    [readout] =&gt; Array
        (
            [slug] =&gt; readout
            [parent] =&gt; 
            [name] =&gt; readout
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; 
        )

    [reasoning] =&gt; Array
        (
            [slug] =&gt; reasoning
            [parent] =&gt; 
            [name] =&gt; reasoning
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; 
            [subentries] =&gt; Array
                (
                    [reasoning!abductive] =&gt; Array
                        (
                            [slug] =&gt; reasoning!abductive
                            [parent] =&gt; reasoning
                            [name] =&gt; abductive
                            [canonical] =&gt; 
                            [see] =&gt; abductive reasoning
                            [description] =&gt; 
                        )

                    [reasoning!analogy] =&gt; Array
                        (
                            [slug] =&gt; reasoning!analogy
                            [parent] =&gt; reasoning
                            [name] =&gt; analogy
                            [canonical] =&gt; 
                            [see] =&gt; reasoning by analogy
                            [description] =&gt; 
                        )

                    [reasoning!antecedent-driven] =&gt; Array
                        (
                            [slug] =&gt; reasoning!antecedent-driven
                            [parent] =&gt; reasoning
                            [name] =&gt; antecedent-driven
                            [canonical] =&gt; 
                            [see] =&gt; backward reasoning
                            [description] =&gt; 
                        )

                    [reasoning!backward] =&gt; Array
                        (
                            [slug] =&gt; reasoning!backward
                            [parent] =&gt; reasoning
                            [name] =&gt; backward
                            [canonical] =&gt; 
                            [see] =&gt; backward reasoning
                            [description] =&gt; 
                        )

                    [reasoning!bayesian] =&gt; Array
                        (
                            [slug] =&gt; reasoning!bayesian
                            [parent] =&gt; reasoning
                            [name] =&gt; Bayesian
                            [canonical] =&gt; 
                            [see] =&gt; bayesian reasoning
                            [description] =&gt; 
                        )

                    [reasoning!bottom-up] =&gt; Array
                        (
                            [slug] =&gt; reasoning!bottom-up
                            [parent] =&gt; reasoning
                            [name] =&gt; bottom-up
                            [canonical] =&gt; 
                            [see] =&gt; 
                            [description] =&gt; 
                        )

                    [reasoning!case-based] =&gt; Array
                        (
                            [slug] =&gt; reasoning!case-based
                            [parent] =&gt; reasoning
                            [name] =&gt; case-based
                            [canonical] =&gt; 
                            [see] =&gt; case-based reasoning
                            [description] =&gt; 
                        )

                    [reasoning!certainty factors] =&gt; Array
                        (
                            [slug] =&gt; reasoning!certainty factors
                            [parent] =&gt; reasoning
                            [name] =&gt; certainty factors
                            [canonical] =&gt; 
                            [see] =&gt; certainty factors
                            [description] =&gt; 
                        )

                    [reasoning!consequent-driven] =&gt; Array
                        (
                            [slug] =&gt; reasoning!consequent-driven
                            [parent] =&gt; reasoning
                            [name] =&gt; consequent-driven
                            [canonical] =&gt; 
                            [see] =&gt; backward reasoning
                            [description] =&gt; 
                        )

                    [reasoning!data-driven] =&gt; Array
                        (
                            [slug] =&gt; reasoning!data-driven
                            [parent] =&gt; reasoning
                            [name] =&gt; data-driven
                            [canonical] =&gt; 
                            [see] =&gt; forward reasoning
                            [description] =&gt; 
                        )

                    [reasoning!deductive] =&gt; Array
                        (
                            [slug] =&gt; reasoning!deductive
                            [parent] =&gt; reasoning
                            [name] =&gt; deductive
                            [canonical] =&gt; 
                            [see] =&gt; deductive reasoning
                            [description] =&gt; 
                        )

                    [reasoning!forward] =&gt; Array
                        (
                            [slug] =&gt; reasoning!forward
                            [parent] =&gt; reasoning
                            [name] =&gt; forward
                            [canonical] =&gt; 
                            [see] =&gt; forward reasoning
                            [description] =&gt; 
                        )

                    [reasoning!fuzzy] =&gt; Array
                        (
                            [slug] =&gt; reasoning!fuzzy
                            [parent] =&gt; reasoning
                            [name] =&gt; fuzzy
                            [canonical] =&gt; 
                            [see] =&gt; fuzzy reasoning
                            [description] =&gt; 
                        )

                    [reasoning!goal-driven] =&gt; Array
                        (
                            [slug] =&gt; reasoning!goal-driven
                            [parent] =&gt; reasoning
                            [name] =&gt; goal-driven
                            [canonical] =&gt; 
                            [see] =&gt; backward reasoning
                            [description] =&gt; 
                        )

                    [reasoning!hypothesis-driven] =&gt; Array
                        (
                            [slug] =&gt; reasoning!hypothesis-driven
                            [parent] =&gt; reasoning
                            [name] =&gt; hypothesis-driven
                            [canonical] =&gt; 
                            [see] =&gt; hypothesis-driven reasoning
                            [description] =&gt; 
                        )

                    [reasoning!inductive] =&gt; Array
                        (
                            [slug] =&gt; reasoning!inductive
                            [parent] =&gt; reasoning
                            [name] =&gt; inductive
                            [canonical] =&gt; 
                            [see] =&gt; inductive reasoning
                            [description] =&gt; 
                        )

                    [reasoning!monotonic] =&gt; Array
                        (
                            [slug] =&gt; reasoning!monotonic
                            [parent] =&gt; reasoning
                            [name] =&gt; monotonic
                            [canonical] =&gt; 
                            [see] =&gt; monotonic reasoning
                            [description] =&gt; 
                        )

                    [reasoning!non-monotonic] =&gt; Array
                        (
                            [slug] =&gt; reasoning!non-monotonic
                            [parent] =&gt; reasoning
                            [name] =&gt; non-monotonic
                            [canonical] =&gt; 
                            [see] =&gt; non-monotonic reasoning
                            [description] =&gt; 
                        )

                    [reasoning!probabilistic] =&gt; Array
                        (
                            [slug] =&gt; reasoning!probabilistic
                            [parent] =&gt; reasoning
                            [name] =&gt; probabilistic
                            [canonical] =&gt; 
                            [see] =&gt; probabilistic reasoning
                            [description] =&gt; 
                        )

                    [reasoning!top-down] =&gt; Array
                        (
                            [slug] =&gt; reasoning!top-down
                            [parent] =&gt; reasoning
                            [name] =&gt; top-down
                            [canonical] =&gt; 
                            [see] =&gt; top-down reasoning
                            [description] =&gt; 
                        )

                )

        )

    [reasoning!abductive] =&gt; Array
        (
            [slug] =&gt; reasoning!abductive
            [parent] =&gt; reasoning
            [name] =&gt; abductive
            [canonical] =&gt; 
            [see] =&gt; abductive reasoning
            [description] =&gt; 
        )

    [reasoning!analogy] =&gt; Array
        (
            [slug] =&gt; reasoning!analogy
            [parent] =&gt; reasoning
            [name] =&gt; analogy
            [canonical] =&gt; 
            [see] =&gt; reasoning by analogy
            [description] =&gt; 
        )

    [reasoning!antecedent-driven] =&gt; Array
        (
            [slug] =&gt; reasoning!antecedent-driven
            [parent] =&gt; reasoning
            [name] =&gt; antecedent-driven
            [canonical] =&gt; 
            [see] =&gt; backward reasoning
            [description] =&gt; 
        )

    [reasoning!backward] =&gt; Array
        (
            [slug] =&gt; reasoning!backward
            [parent] =&gt; reasoning
            [name] =&gt; backward
            [canonical] =&gt; 
            [see] =&gt; backward reasoning
            [description] =&gt; 
        )

    [reasoning!backward chaining] =&gt; Array
        (
            [slug] =&gt; reasoning!backward chaining
            [parent] =&gt; reasoning
            [name] =&gt; backward chaining
            [canonical] =&gt; reasoning!backward
            [see] =&gt; 
            [description] =&gt; 
        )

    [reasoning!bayesian] =&gt; Array
        (
            [slug] =&gt; reasoning!bayesian
            [parent] =&gt; reasoning
            [name] =&gt; Bayesian
            [canonical] =&gt; 
            [see] =&gt; bayesian reasoning
            [description] =&gt; 
        )

    [reasoning!bottom-up] =&gt; Array
        (
            [slug] =&gt; reasoning!bottom-up
            [parent] =&gt; reasoning
            [name] =&gt; bottom-up
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; 
        )

    [reasoning!case-based] =&gt; Array
        (
            [slug] =&gt; reasoning!case-based
            [parent] =&gt; reasoning
            [name] =&gt; case-based
            [canonical] =&gt; 
            [see] =&gt; case-based reasoning
            [description] =&gt; 
        )

    [reasoning!certainty factors] =&gt; Array
        (
            [slug] =&gt; reasoning!certainty factors
            [parent] =&gt; reasoning
            [name] =&gt; certainty factors
            [canonical] =&gt; 
            [see] =&gt; certainty factors
            [description] =&gt; 
        )

    [reasoning!consequent-driven] =&gt; Array
        (
            [slug] =&gt; reasoning!consequent-driven
            [parent] =&gt; reasoning
            [name] =&gt; consequent-driven
            [canonical] =&gt; 
            [see] =&gt; backward reasoning
            [description] =&gt; 
        )

    [reasoning!data-driven] =&gt; Array
        (
            [slug] =&gt; reasoning!data-driven
            [parent] =&gt; reasoning
            [name] =&gt; data-driven
            [canonical] =&gt; 
            [see] =&gt; forward reasoning
            [description] =&gt; 
        )

    [reasoning!deductive] =&gt; Array
        (
            [slug] =&gt; reasoning!deductive
            [parent] =&gt; reasoning
            [name] =&gt; deductive
            [canonical] =&gt; 
            [see] =&gt; deductive reasoning
            [description] =&gt; 
        )

    [reasoning!forward] =&gt; Array
        (
            [slug] =&gt; reasoning!forward
            [parent] =&gt; reasoning
            [name] =&gt; forward
            [canonical] =&gt; 
            [see] =&gt; forward reasoning
            [description] =&gt; 
        )

    [reasoning!forward chaining] =&gt; Array
        (
            [slug] =&gt; reasoning!forward chaining
            [parent] =&gt; reasoning
            [name] =&gt; forward chaining
            [canonical] =&gt; reasoning!forward
            [see] =&gt; 
            [description] =&gt; 
        )

    [reasoning!fuzzy] =&gt; Array
        (
            [slug] =&gt; reasoning!fuzzy
            [parent] =&gt; reasoning
            [name] =&gt; fuzzy
            [canonical] =&gt; 
            [see] =&gt; fuzzy reasoning
            [description] =&gt; 
        )

    [reasoning!goal-driven] =&gt; Array
        (
            [slug] =&gt; reasoning!goal-driven
            [parent] =&gt; reasoning
            [name] =&gt; goal-driven
            [canonical] =&gt; 
            [see] =&gt; backward reasoning
            [description] =&gt; 
        )

    [reasoning!hypothesis-driven] =&gt; Array
        (
            [slug] =&gt; reasoning!hypothesis-driven
            [parent] =&gt; reasoning
            [name] =&gt; hypothesis-driven
            [canonical] =&gt; 
            [see] =&gt; hypothesis-driven reasoning
            [description] =&gt; 
        )

    [reasoning!inductive] =&gt; Array
        (
            [slug] =&gt; reasoning!inductive
            [parent] =&gt; reasoning
            [name] =&gt; inductive
            [canonical] =&gt; 
            [see] =&gt; inductive reasoning
            [description] =&gt; 
        )

    [reasoning!monotonic] =&gt; Array
        (
            [slug] =&gt; reasoning!monotonic
            [parent] =&gt; reasoning
            [name] =&gt; monotonic
            [canonical] =&gt; 
            [see] =&gt; monotonic reasoning
            [description] =&gt; 
        )

    [reasoning!non-monotonic] =&gt; Array
        (
            [slug] =&gt; reasoning!non-monotonic
            [parent] =&gt; reasoning
            [name] =&gt; non-monotonic
            [canonical] =&gt; 
            [see] =&gt; non-monotonic reasoning
            [description] =&gt; 
        )

    [reasoning!probabilistic] =&gt; Array
        (
            [slug] =&gt; reasoning!probabilistic
            [parent] =&gt; reasoning
            [name] =&gt; probabilistic
            [canonical] =&gt; 
            [see] =&gt; probabilistic reasoning
            [description] =&gt; 
        )

    [reasoning!top-down] =&gt; Array
        (
            [slug] =&gt; reasoning!top-down
            [parent] =&gt; reasoning
            [name] =&gt; top-down
            [canonical] =&gt; 
            [see] =&gt; top-down reasoning
            [description] =&gt; 
        )

    [reasoning by analogy] =&gt; Array
        (
            [slug] =&gt; reasoning by analogy
            [parent] =&gt; 
            [name] =&gt; reasoning by analogy
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; Analogies are used extensively in {{human reasoning}} to help to understand new situations and to suggest potential problems solutions, for example, given the analogy &quot; a school teacher with their pupils are like a shepherd herding sheep&quot;, one might reason that as shepherds are better able to direct sheep is they are kept in a group, this might be a good strtegy for schoolteachers too.  Analogies can also be used by AI systems by noting structural similarities between past cases and the current situation.  As the analogy may fail to hold at some point, analogical reasoning may be used as a {{heuristic}}.  Within AI, the uses of AI are usually either as a {{transformational analogy}} (using at solutions to analogous problems) or a {{derivational analogy}} (using the process of obtaining the analogous solution).  Reasoning by analogy is often alongside {{case-based reasoning}}.
        )

    [reasoning systems] =&gt; Array
        (
            [slug] =&gt; reasoning systems
            [parent] =&gt; 
            [name] =&gt; reasoning systems
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; 
        )

    [reasoning with certainty factors] =&gt; Array
        (
            [slug] =&gt; reasoning with certainty factors
            [parent] =&gt; 
            [name] =&gt; reasoning with certainty factors
            [canonical] =&gt; 
            [see] =&gt; certainty factors
            [description] =&gt; 
        )

    [recall] =&gt; Array
        (
            [slug] =&gt; recall
            [parent] =&gt; 
            [name] =&gt; recall
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; When making a binary classification or decision, such as a disease diagnosis, recall is the {{probability}} that if the person really has the disease, or in general if the classification really is positive, then we  corectly make the decision, for example, if the person has the disease then we do diagnose it.   If a decision has a high recall, then the probability of a {{false negative}} is low.  Often increasing recall reduces {{precision}}; this is called the {{precision-recall trade-off}}.
        )

    [recaptcha] =&gt; Array
        (
            [slug] =&gt; recaptcha
            [parent] =&gt; 
            [name] =&gt; reCAPTCHA
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; reCAPTCHA is a common security feature on web pages used to distinguish human users from automated {{bots}}.  reCAPTCHA appears as a small dialogue or popup on a website asking the user to perform a simple tasks such as read text or match images.  The tasks are designed to be simple for humans, but hard for machines, hence useful to distinguish then, indeed the term CAPTCHA is originally an acronym for &quot;Completely Automated Public {{Turing test}} to tell Computers and Humans Apart&quot;.  While this is their primary purpose, they are also used to  fill in gaps in AI algorithms and to create {{labelled training data}} for {{machine learning}}.  The original reCAPTCHA helped fill in difficult portions of OCR for scanned books and is an early example of {{human computation}}.
        )

    [recaptcha codes] =&gt; Array
        (
            [slug] =&gt; recaptcha codes
            [parent] =&gt; 
            [name] =&gt; recaptcha codes
            [canonical] =&gt; recaptcha
            [see] =&gt; 
            [description] =&gt; 
        )

    [receiver operating characteristic] =&gt; Array
        (
            [slug] =&gt; receiver operating characteristic
            [parent] =&gt; 
            [name] =&gt; receiver operating characteristic
            [canonical] =&gt; 
            [see] =&gt; roc
            [description] =&gt; 
        )

    [receptive field] =&gt; Array
        (
            [slug] =&gt; receptive field
            [parent] =&gt; 
            [name] =&gt; receptive field
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; 
        )

    [recidivism] =&gt; Array
        (
            [slug] =&gt; recidivism
            [parent] =&gt; 
            [name] =&gt; recidivism
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; Recidivism is a technical term in the judicial system refering to the likelihood that soemone who has previsouly committed. acrime will do so again.  The controversial AI system {{COMPAS}} creates a precdicted recidivism score (effectively high or low likelhood of reoffending) and has been used by judges in the USA as part of parole decisions.  {{COMPAS}} was built using {{machine learning}} over past felons and their actual reoffending rates, so in a way is based on {{ground truth}} data.  However, in practice the scores appear to exhibit significant {{racial bias}}, made worse because {{automation bias}} means the scores may effectively be treated as recommendations.
        )

    [recommender systems] =&gt; Array
        (
            [slug] =&gt; recommender systems
            [parent] =&gt; 
            [name] =&gt; recommender systems
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; Recommender systems use data collected about past behaviours, such as books or music choices, of large numbers of people and use this to suggest new things to read, or listen to.  They use {{bog data}} echnques such as lareg-scale {{statistical methods}}.
        )

    [recurrent neural network] =&gt; Array
        (
            [slug] =&gt; recurrent neural network
            [parent] =&gt; 
            [name] =&gt; recurrent neural network
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; A recurrent neural network is a form of {{neural network}} with a state/memory that runs on {{time series}} or {{sequential}} data.  At each step a {{neura networ}} uses the current state and inout token to geerate the next state and to predict the next token.  As the hidden state is not part of the training data, the {{training phase}} is more complex than for a simple {{neural network}}.
        )

    [recurrent neural networks] =&gt; Array
        (
            [slug] =&gt; recurrent neural networks
            [parent] =&gt; 
            [name] =&gt; recurrent neural networks
            [canonical] =&gt; recurrent neural network
            [see] =&gt; 
            [description] =&gt; 
        )

    [reduce] =&gt; Array
        (
            [slug] =&gt; reduce
            [parent] =&gt; 
            [name] =&gt; reduce
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; Reduce is a standard function in {{Lisp}} and other {{functional programming}} languages.  The basic reduce function has two arguments reduce(f,list), where the function f takes two arguments and is initially applied to the first two elements in the list, then the fucntion is applied again to the result and the third item, and so on. For example: &lt;br&gt; &amp;nbsp; &amp;nbsp; &amp;nbsp; reduce( '+', (1,2,3,4) )  =  (((1+2)+3)+4). &lt;br&gt; Reduce functions typically allow an optional third argument that is used as an initial value. &lt;br&gt;&lt;br&gt; The reduce function is one of the inspirations for the {{MapReduce}} infrastructure for massive scale processing of {{big data}}.
        )

    [reduce!lisp] =&gt; Array
        (
            [slug] =&gt; reduce!lisp
            [parent] =&gt; reduce
            [name] =&gt; lisp
            [canonical] =&gt; reduce
            [see] =&gt; 
            [description] =&gt; 
        )

    [redundancy] =&gt; Array
        (
            [slug] =&gt; redundancy
            [parent] =&gt; 
            [name] =&gt; redundancy
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; Redundancy is when sensor readings or items of data replicate in full or part data that is sensed or stored elsewhere.  Redundancy can be seen as a source of {{inefficiency}} so that one may try to remove redundant values as part of {{data clearening}}.  However, redundancy can also be critical in {{fault tolerant}} systems allowinga. level of recovery if a sensor fails or data is lost.  Redundancy may be precise, in that one peoce of data can be calculated from others.  Alterbatively, the redundant data items may be statistically related, for example where several values in some way measure or represnt the same thing, but have a degree of noise.  In  this case {{statistical}} techniques, including simple averaging or {[time series}} techniques can be used to obtian a smaller number of more accurate readings.  In {{actvity recognition}} systems redundant sensors give more {[robust}} detection, but also mean that as new sensors are addedd, they can be {{correlated}} with existing sensors, adding to redundacy especially if existing sensors fail or are removed and allowing deployments of sensors to change whilst retaining higher-level reasoning.
        )

    [redundant data] =&gt; Array
        (
            [slug] =&gt; redundant data
            [parent] =&gt; 
            [name] =&gt; redundant data
            [canonical] =&gt; redundancy
            [see] =&gt; 
            [description] =&gt; 
        )

    [region detection] =&gt; Array
        (
            [slug] =&gt; region detection
            [parent] =&gt; 
            [name] =&gt; region detection
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; 
        )

    [region growing] =&gt; Array
        (
            [slug] =&gt; region growing
            [parent] =&gt; 
            [name] =&gt; region growing
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; 
        )

    [regression line] =&gt; Array
        (
            [slug] =&gt; regression line
            [parent] =&gt; 
            [name] =&gt; regression line
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; In {{linear regression}}, the {{regression line}} is the line that best fits the data.
        )

    [regret] =&gt; Array
        (
            [slug] =&gt; regret
            [parent] =&gt; 
            [name] =&gt; regret
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; 
        )

    [reinforcement] =&gt; Array
        (
            [slug] =&gt; reinforcement
            [parent] =&gt; 
            [name] =&gt; reinforcement
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; Reinforcement is the process of increasing or decreasing the weightings or parameters that gave rise to a positive or negative  behaviour.  The idea is based on Skinner's experiments with pigeons in the 1950s.  If the mapoing between stimulus, behaviour and reward is simple and direct, for example a {{look-up table}} with weights, then reinforcement is relatuveky straughtfirward.  More commonly however, the {{causal relatiinship}} is complex and there are problems of {{credit assignment}}.
        )

    [reinforcement learning] =&gt; Array
        (
            [slug] =&gt; reinforcement learning
            [parent] =&gt; 
            [name] =&gt; reinforcement learning
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; Reinforcement learning is used to learn complex time and situation behaviours, when there is eiether no training data or it is insufficient.  Often the poitn ar which rewards occur are well after the action that caused them and may be the result of several past actions leading toe {{credit assignment}} problems, which make positive or negative {{reinforcement}} hard.  Learning takes place due to interactions with a (real ir simukated) world and therfore have a cost both directly due to the action being perfroemd (energy expenditure ofr a robot, network costs for a web agent) and indirecly due to the posutuve or negative effects of the action. However, without taking actions there is no potential for learning, this leads to an {{exploration-exploitation trade-off}}.
        )

    [relational database] =&gt; Array
        (
            [slug] =&gt; relational database
            [parent] =&gt; 
            [name] =&gt; relational database
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; 
        )

    [relaxation term] =&gt; Array
        (
            [slug] =&gt; relaxation term
            [parent] =&gt; 
            [name] =&gt; relaxation term
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; A relaxation term is used to govern the speen of adadptation of many kinds of {{iterative algorithm}}.  These algorithms are often have the form: &lt;br&gt; &amp;nbsp; &amp;nbsp; &amp;nbsp; new_value = old_value + &amp;lambda; change_direction. &lt;br&gt; The term &amp;lambda; is the relaxation term.  If this is too large the algorithm is {{unstable}} swinging wildly between different values, if it is too small the algorithm takes a very ling time to {{converge}}.  In some algoruthms, notably {{simulated annealing}} the relaxation terms starts relatively large (but not so large to cause instability) to allow rapid movement towards the general site of a solutuon, and then gradually reduced to allow fine tuning.
        )

    [relevance feedback] =&gt; Array
        (
            [slug] =&gt; relevance feedback
            [parent] =&gt; 
            [name] =&gt; relevance feedback
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; Relevance feedback is used in {{docuemnt retreival}}, {{recommender systes}} and other forms of {{intelligent user interfaces}} to allow users to say how good autimated suggestions are.  This is then fed back into the algorithms to improve future suggestions.  In some cases relevance feedback is {[explicit}}, for example, adding a star rating or thumbs up, in others it is {{implicit}} for example, if the users clicks through for more information about a suggested item.  {{Epistemic interaction}} can be used to design interactions that increase this feedback.
        )

    [relevance information] =&gt; Array
        (
            [slug] =&gt; relevance information
            [parent] =&gt; 
            [name] =&gt; relevance information
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; {{Web search}} systems may use various kinds of relevance information as well as the presence of gthe search terms in text.  For example, if search words occur close to one another, or if page linking to a specific page use the terms as their link text.
        )

    [reparameterisation] =&gt; Array
        (
            [slug] =&gt; reparameterisation
            [parent] =&gt; 
            [name] =&gt; reparameterisation
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; Reparamerisation is where one transforms the parameters that describe a problem or solution space in such a way that preserves information, but may be easier to process or learn from.  For example, changing a date represented as DD/MM/YYY into seconds since 1970 in order to facilitate rapid comparisons.
        )

    [res cogitans] =&gt; Array
        (
            [slug] =&gt; res cogitans
            [parent] =&gt; 
            [name] =&gt; res cogitans
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; Res cogitans is the term used by {{Descartes}} to refer to mental stuff, which he regarded as different from {{res extensa}} the physical material of our bodies.  This is a form of {{dualism}}, which is rejected by most modern philosophers ... except insofar as many feel that biological brains are intrinsically different from computers.
        )

    [res extensa] =&gt; Array
        (
            [slug] =&gt; res extensa
            [parent] =&gt; 
            [name] =&gt; res extensa
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; Res extensa is the term used by {{Descartes}} to refer to the physical material of our bodies, which he regarded as different from {{res cogitans}} the mental stuff that allows us to think.  This is a form of {{dualism}}, which is rejected by most modern philosophers ... except insofar as many feel that biological brains are intrinsically different from computers.
        )

    [rescue dogs] =&gt; Array
        (
            [slug] =&gt; rescue dogs
            [parent] =&gt; 
            [name] =&gt; rescue dogs
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; Within the book 'rescue dogs' refers to search and rescue, for example the St Bernard used in mountain rescue.  However, the term can also mean dogs that are waiting to be re-homed.
        )

    [responsibility] =&gt; Array
        (
            [slug] =&gt; responsibility
            [parent] =&gt; 
            [name] =&gt; responsibility
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; 
        )

    [reservoir computing] =&gt; Array
        (
            [slug] =&gt; reservoir computing
            [parent] =&gt; 
            [name] =&gt; reservoir computing
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; Reservoir computing uses a variety of physical, biologcal or computational processes to increase the {{non-linear diversity}} of input data the outputs of whih can then be used as inputs for a simpler final {{machine learning}} stage..  For example, input data might be used to drive electrical impolses into a semi-chaotic silicon substrate and the output currents measured at multiple points.
        )

    [residual] =&gt; Array
        (
            [slug] =&gt; residual
            [parent] =&gt; 
            [name] =&gt; residual
            [canonical] =&gt; residuals
            [see] =&gt; 
            [description] =&gt; 
        )

    [residuals] =&gt; Array
        (
            [slug] =&gt; residuals
            [parent] =&gt; 
            [name] =&gt; residuals
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; The residuals are the difference between actual data values and their modelled values after some sort of {{data fitting}} process.  For example in {{linear regression}} the distance of the data points from the {{regression line}}.  In general, for data of the form (x&lt;sub&gt;i&lt;/sub&gt;,y(sub&gt;i&lt;/sub&gt;) if a model 'M' has been fitted to predict y form x, then the residuals r&lt;sub&gt;i&lt;/sub&gt; are given by: &lt;br&gt; &amp;nbsp; &amp;nbsp; &amp;nbsp; r&lt;sub&gt;i&lt;/sub&gt; &amp;nbsp; = &amp;nbsp; y(sub&gt;i&lt;/sub&gt; &ndash; M(x&lt;sub&gt;i&lt;/sub&gt;) &lt;br&gt; Often the model fitting process is expressed in reducing the {[sum of squares}} of the residuals or some similar metric as a {{fitness function}}.
        )

    [resource description framework] =&gt; Array
        (
            [slug] =&gt; resource description framework
            [parent] =&gt; 
            [name] =&gt; resource description framework
            [canonical] =&gt; 
            [see] =&gt; rdf
            [description] =&gt; 
        )

    [resource description framework schema] =&gt; Array
        (
            [slug] =&gt; resource description framework schema
            [parent] =&gt; 
            [name] =&gt; resource description framework schema
            [canonical] =&gt; 
            [see] =&gt; rdfs
            [description] =&gt; 
        )

    [responsible innovation] =&gt; Array
        (
            [slug] =&gt; responsible innovation
            [parent] =&gt; 
            [name] =&gt; responsible innovation
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; Responsible innovaton (RI) means understanding the potential ethical implications of one's work including both societal and environmental impacts and both the outcomes and the process by which it is performed.  For example, you may be developing a new system for a charity (ethically good outcome), but not think about the impact on the participants of the initial prototype when it is closed down.  The impacts may also be indirect or future for example, developing systems to improve urban traffic flow might have short term postive impacts, but lead to more car use in the longer term.  These issues apply both the product deelopment and reesearch.  For the latter one often has to imagine the longer term trajectory of the research -- the UK research funding agency EPSRC have developed the {{AREA framework}} to help researchers in the RI process.
        )

    [restricted boltzmann machine] =&gt; Array
        (
            [slug] =&gt; restricted boltzmann machine
            [parent] =&gt; 
            [name] =&gt; Restricted Boltzmann Machine
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; A Restricted Boltzmann Machine (RBM) is a form of {{Boltzmann Machine}} where input and output layers are fully connected to each another, but have no internal connections.  That is gthere are no input-input, not output-output links.  Otherwsie they are like a standard {{Boltzmann Machine} } with a training phase with inputs clamped and outputs free to vary and then a recall mode where some inputs are clamped and the network fills in the gaps.  As well as this reocnstreuction mode, RBMs are often used for {{feature detction}} in the early stages of {{deep neaural networks}} as they use feedforward learning rather than {{backpropagation}}.      The Restricted Boltzmann Machine (RBM) like the {{Boltzmann Machine}} is a form of {{neural network}} and {{autoencoder}}.
        )

    [restricted boltzmann machines] =&gt; Array
        (
            [slug] =&gt; restricted boltzmann machines
            [parent] =&gt; 
            [name] =&gt; restricted boltzmann machines
            [canonical] =&gt; restricted boltzmann machine
            [see] =&gt; 
            [description] =&gt; 
        )

    [ridge] =&gt; Array
        (
            [slug] =&gt; ridge
            [parent] =&gt; 
            [name] =&gt; ridge
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; A ridge in the {{fitness landscape}} is when there is a direction of change in the parameters that makes very little difference to the {{fitness function}}, but where small changes in any other direction drop the fitness radically.  For example, suppose the fitness in an area is approximately 1&amp;minus;&amp;radic;| x&amp;minus;y | and we start at the (x,y) point (2,2) where the fitness function is zero.  Small changes to (2+&amp;delta;&lt;sub&gt;x&lt;/sub&gt;, 2+&amp;delta;&lt;sub&gt;y&lt;/sub&gt;) stay at one if &amp;delta;&lt;sub&gt;x&lt;/sub&gt;=&amp;delta;&lt;sub&gt;y&lt;/sub&gt;, but if they are not equal even quite small changes where &amp;delta;&lt;sub&gt;x&lt;/sub&gt;&amp;ne;&amp;delta;&lt;sub&gt;y&lt;/sub&gt; will lead to rapid drops in the fitness function;
        )

    [risk avoidance] =&gt; Array
        (
            [slug] =&gt; risk avoidance
            [parent] =&gt; 
            [name] =&gt; risk avoidance
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; Risk avoidance is where one deiberately adopts a strtegy hat reduces {{risk}}, for example prefering solutions in a {{gam space}} that have lower expected payback, but less variability.  For exampel, the {{minimax algotihm}} is effectively about minimising the worst case loss against a perfect player.  A less {{risk averse}} approach against an imperfect opponent might be to choose a path that has a small potential to lead to a bad defeat, but has also many ways once can turn the game around and win.  {{Stochastic games}} where there is an element of chance may mean an AI player has to choose between options that have low potential yield, but also low {{uncertainty}}, and those with higher expected gains,but with a lot of {{uncertainty}}.  By  adjusting the balance between these one can make the asystem more or less {{risk averse}}. 
        )

    [rnn] =&gt; Array
        (
            [slug] =&gt; rnn
            [parent] =&gt; 
            [name] =&gt; RNN
            [canonical] =&gt; 
            [see] =&gt; recurrent neural network
            [description] =&gt; 
        )

    [robert wadlow] =&gt; Array
        (
            [slug] =&gt; robert wadlow
            [parent] =&gt; 
            [name] =&gt; Robert Wadlow
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; Robert Wadlow was the tallest person ever recorded, standing 2.72 metres tall (8ft 11inches).  He is an example of an {{outlier}} that is real, albeit unusual, data rather than an error in data collection or processing.
        )

    [robert's operator] =&gt; Array
        (
            [slug] =&gt; robert's operator
            [parent] =&gt; 
            [name] =&gt; Robert's operator
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; Robert's operator is a {{non-linear}} filter to detect the presence of lines in an image. It work son a ecah 2x2 portion of the scence and is defined as the sum of the absolute differences between opposing corners: &lt;br&gt; &amp;nbsp; &amp;nbsp; &amp;nbsp;  G(x,y)  &amp;nbsp;  =  &amp;nbsp;  | f(x,y) &amp;minus; f(x+1,y-1)  | &amp;nbsp; + &amp;nbsp;  | f(x+1,y) &amp;minus;  f(x,y-1) |$
        )

    [robotics] =&gt; Array
        (
            [slug] =&gt; robotics
            [parent] =&gt; 
            [name] =&gt; robotics
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; Robotics is the arae of AI and engineering that studies the physical construction and computational control of {{robots}}.
            [subentries] =&gt; Array
                (
                    [robotics!current state representation] =&gt; Array
                        (
                            [slug] =&gt; robotics!current state representation
                            [parent] =&gt; robotics
                            [name] =&gt; current state representation
                            [canonical] =&gt; 
                            [see] =&gt; current state representation
                            [description] =&gt; 
                        )

                    [robotics!feedback] =&gt; Array
                        (
                            [slug] =&gt; robotics!feedback
                            [parent] =&gt; robotics
                            [name] =&gt; feedback
                            [canonical] =&gt; 
                            [see] =&gt; 
                            [description] =&gt; 
                        )

                    [robotics!historical representation] =&gt; Array
                        (
                            [slug] =&gt; robotics!historical representation
                            [parent] =&gt; robotics
                            [name] =&gt; historical representation
                            [canonical] =&gt; 
                            [see] =&gt; historical representation
                            [description] =&gt; 
                        )

                    [robotics!practical] =&gt; Array
                        (
                            [slug] =&gt; robotics!practical
                            [parent] =&gt; robotics
                            [name] =&gt; practical
                            [canonical] =&gt; 
                            [see] =&gt; 
                            [description] =&gt; 
                        )

                    [robotics!safety] =&gt; Array
                        (
                            [slug] =&gt; robotics!safety
                            [parent] =&gt; robotics
                            [name] =&gt; safety
                            [canonical] =&gt; 
                            [see] =&gt; 
                            [description] =&gt; 
                        )

                )

        )

    [robotics!current state representation] =&gt; Array
        (
            [slug] =&gt; robotics!current state representation
            [parent] =&gt; robotics
            [name] =&gt; current state representation
            [canonical] =&gt; 
            [see] =&gt; current state representation
            [description] =&gt; 
        )

    [robotics!feedback] =&gt; Array
        (
            [slug] =&gt; robotics!feedback
            [parent] =&gt; robotics
            [name] =&gt; feedback
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; 
        )

    [robotics!historical representation] =&gt; Array
        (
            [slug] =&gt; robotics!historical representation
            [parent] =&gt; robotics
            [name] =&gt; historical representation
            [canonical] =&gt; 
            [see] =&gt; historical representation
            [description] =&gt; 
        )

    [robotics!practical] =&gt; Array
        (
            [slug] =&gt; robotics!practical
            [parent] =&gt; robotics
            [name] =&gt; practical
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; 
        )

    [robotics!safety] =&gt; Array
        (
            [slug] =&gt; robotics!safety
            [parent] =&gt; robotics
            [name] =&gt; safety
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; 
        )

    [robots] =&gt; Array
        (
            [slug] =&gt; robots
            [parent] =&gt; 
            [name] =&gt; robots
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; 
        )

    [robust to failure] =&gt; Array
        (
            [slug] =&gt; robust to failure
            [parent] =&gt; 
            [name] =&gt; robust to failure
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; 
        )

    [roc] =&gt; Array
        (
            [slug] =&gt; roc
            [parent] =&gt; 
            [name] =&gt; ROC
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; A ROC diagram or ROC curve is a visualisation of the {{precision-recall trade-off}}.    A ROC diagram can also be used to help choise between classifiers. 
        )

    [roc curve] =&gt; Array
        (
            [slug] =&gt; roc curve
            [parent] =&gt; 
            [name] =&gt; ROC curve
            [canonical] =&gt; roc
            [see] =&gt; 
            [description] =&gt; 
        )

    [ross quinlan] =&gt; Array
        (
            [slug] =&gt; ross quinlan
            [parent] =&gt; 
            [name] =&gt; Ross Quinlan
            [canonical] =&gt; 
            [see] =&gt; quinlan, ross
            [description] =&gt; 
        )

    [rote learning] =&gt; Array
        (
            [slug] =&gt; rote learning
            [parent] =&gt; 
            [name] =&gt; rote learning
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; In rote learning one learns precisely the examples given.  For humans this can be hard, but for a machine it is easy, just store all training examples with ther exepcted output and then lookup the correct value.   On its own this cannot {{generalize}}, when faced with a novel input that it has not seen before it has no way of filling in or estimating the correct output based on the inputs.  While this is a very simple form of {{machine learning}}, it can be a valuable part of more complex learning systems.
        )

    [rote learning!memorizing] =&gt; Array
        (
            [slug] =&gt; rote learning!memorizing
            [parent] =&gt; rote learning
            [name] =&gt; memorizing
            [canonical] =&gt; rote learning
            [see] =&gt; 
            [description] =&gt; 
        )

    [route planning] =&gt; Array
        (
            [slug] =&gt; route planning
            [parent] =&gt; 
            [name] =&gt; route planning
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; Route planning is about finding paths form one location to another usually restructed to particular kinds of connection (such as roads or railways) and often with some form of {{optimisation criterion}} (such as minimal fuel costs) and {{constraints}} (such as passing particular intermediate points, or avoiding narrow roads).  If all of the distances, links and locatiosn are known then various forms of graph search can be used, notably the {{A* algorithm}}.  Where some parts are not known in advance there may need to be some form of high-level global planning to determine a best-guess approxiomate route, with local planning if problems are encountered, such as.a blocked road.
        )

    [rubin's vase] =&gt; Array
        (
            [slug] =&gt; rubin's vase
            [parent] =&gt; 
            [name] =&gt; Rubin's vase
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; Rubin's vase is an {{ambiguous image}}, which can be seen as vase or as two faces looking at one another.  Human viewers tend to see such images flip {{quasi-periodically}} between the two interpretations: vase, then faces, then vase again ... This gives clues as to how the brain deals with ambiguity and hence lessons for computer {{disambiguation}}. 
        )

    [rule induction] =&gt; Array
        (
            [slug] =&gt; rule induction
            [parent] =&gt; 
            [name] =&gt; rule induction
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; 
        )

    [saccades] =&gt; Array
        (
            [slug] =&gt; saccades
            [parent] =&gt; 
            [name] =&gt; saccades
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; Saccades are the raoid movements your eye makes over a scene as you take in the details.  They are needed because a small portion at the center of the retina, called the {{fovea}}, contans the majority of your light senisng cells.    The saccades allows you to build up an overall detailed view rather like taking lots of photographs and then stitching them together to form a panorama. 
        )

    [sail databank] =&gt; Array
        (
            [slug] =&gt; sail databank
            [parent] =&gt; 
            [name] =&gt; SAIL databank
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; The SAIL databank is a large collection of interlinked datasest related to health data promarily in Wales.  It is used for social and health related research..   A central aspect of the datset is that it connects all the relevent data about an indvidual including hosipital visits, GP appointemnts and also data from local government etc.  In order to link new datsets all ingested data is matcehd against a special database of personally identifiable information, and then {{pseudonomised}} so that the unique id for each individual can be matched, but no identifiable information is present in the data saved for future processing.  Even given this {{security}} and {{privacy}} are major issues with stringent access controls and no direct internet connections.
        )

    [salience] =&gt; Array
        (
            [slug] =&gt; salience
            [parent] =&gt; 
            [name] =&gt; salience
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; Salience is the extent to which an event, object, or other stimulus captures {{attention}}.  Typically things have high salience if they (i) stand out visually (or on other senory ways) from their backgorund; (ii) if they are less frequently experienced; or (ii) if they have greater potential impact.  For example, (i) a bright red flower in the middle of a green field; (ii) an ice cream in the middle of winter; or (iii) a lion prowling.  In fact a lion prowling in a shopping street is an example of both (ii) and (iii)!  Salince is crucial to help humans and other animals choose were to focus attention and hence ac and learn more effectively.  Computational models of {{attention}} have been particularly important in the development of {{large language models}}.
        )

    [salient] =&gt; Array
        (
            [slug] =&gt; salient
            [parent] =&gt; 
            [name] =&gt; salient
            [canonical] =&gt; salience
            [see] =&gt; 
            [description] =&gt; 
        )

    [sampling bias] =&gt; Array
        (
            [slug] =&gt; sampling bias
            [parent] =&gt; 
            [name] =&gt; sampling bias
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; Sampling bias occurs when the composition of your sample in some ways skews your results.  For example, if a market researcher goes into a shopping street on a weekday afternoon they will encounter few employed people.  
        )

    [sanity checks] =&gt; Array
        (
            [slug] =&gt; sanity checks
            [parent] =&gt; 
            [name] =&gt; sanity checks
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; Raw data may have errors, mistypings or other problems.  Sanity checks are small, usualy realtively simple, pieces of code run across a dataset to verify simple properties.  For example whether items indexed in a book all appear in the glossary, or whether a field that is intended to be a number indeed only contains numbers.  This is often a critical part of {{data cleaning}} (or {{data wrangling}})   In addition, when dealing with large quantities of data it can be hard to be sure that transformations, which appear sensible in code, actually do what was intended as only a small portion of the data can be inspected by hand.   Sanity checks can again be applied after a transformation stage to help build confidence.  Sanity checs often involve simple {{regular expression}} matching (e.g. to verify the number field) or looking up values (e.g. to check glossary items are defined).
        )

    [satisficing] =&gt; Array
        (
            [slug] =&gt; satisficing
            [parent] =&gt; 
            [name] =&gt; satisficing
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; Satisficing is about acheiving good enough solutions, rtaher than necessarily striving for the very best.  The term was coined by {{Herbert A. Simon}}, the poliical economist and AI pioneer, as a blend between he satisfy and suffice.  {{Human reasoning}} is often driven by satisficing and {{bounded rationality}} as real world decisions often have the made in the absnce of complete information and without time or mental capacity to fully reason through all scenarios.  Although the memory cacpity of machines can allow greater amonts of lookahead, AI systems are ultimately faced with similar situations and this is one of the reasons for the use of {{heuristics}}.
        )

    [scale-related variability] =&gt; Array
        (
            [slug] =&gt; scale-related variability
            [parent] =&gt; 
            [name] =&gt; scale-related variability
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; Scale-related variability is when the variablity, often measured by {{standard deviation}} or {{variance}}) is larger (or smaller) when the underlying value is larger.  For example, if you are measuring times, such the evaporation time of different liquids, it is likely that the variablity of those with long evaporation times is greater than those with fast evaporation.  Often {{log transforms} are used with time-measurements such as these to make the variance more uniform.  This is a form of {{pre-whitening}}.
        )

    [schema.org] =&gt; Array
        (
            [slug] =&gt; schema.org
            [parent] =&gt; 
            [name] =&gt; schema.org
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; Schema.org is a set of standards for embadding semntic information in human-readable web pages. Various mechanisms had emerged as de-facto standards, including {{microformats}}, but aregly for very particular data types and applications.  Several major vendors worked together to create the standards in schema.org, partly to enable better {{web scraping}} by theor own {{crawlers}}.  Scema.org defines {{vocabularies}} of semantic terms (e.g. around personal information, or events) and multiple ways that these can be embedded into web pages {{backwards compatible}} with the earlier emergent standrads. 
        )

    [scientific data] =&gt; Array
        (
            [slug] =&gt; scientific data
            [parent] =&gt; 
            [name] =&gt; scientific data
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; One application of AI, and in particular {{machine learning}}, is to process and interpret scientific data.  This can often be voluminous, indeed the {{world wide web}} was originally developed to help share the vast amunts of data generated at {{CERN}}, it is thus an early example of {{big data}}.  Some forms of processing may simply be to reduce the dataset, ignoring common events or patterns and bringing more interesting ones to the scientists' attention.  However, forms of {{rule induction}} can be used to infer scientific laws directly from data, in a similar way to how Keplar spoted the relationships between planets' orbital distance and speed, namely that they &quot;sweep out equal areas in equal times&quot;.  Automated systems are able to try many different forms of equations and then choose ones that are both simple and fit the data well.
        )

    [screen scraping] =&gt; Array
        (
            [slug] =&gt; screen scraping
            [parent] =&gt; 
            [name] =&gt; screen scraping
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; Screen scraping is the name given to various technologies, tools and pratical techniques that are used to extract information from comouter systems with no built-in data interface or {{API}}.  It is most commoinly applied to old 80x25 terminal-based screen interfaces, but can also be used (with more difficulty) for window-based applications.  Screen scraping can be used purely to obtain information from the current screen, but typically there will be techniques to drive the application, for example emulating key presses or muse clicks.  This can be use to {{crawl}} the application and extract data, or to allow the creation of an alternative (more up-to-date) {{user interface}}.  When screen scraping is applied to web pages it is called {{web scrapping}}.
        )

    [script] =&gt; Array
        (
            [slug] =&gt; script
            [parent] =&gt; 
            [name] =&gt; script
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; 
        )

    [scripts] =&gt; Array
        (
            [slug] =&gt; scripts
            [parent] =&gt; 
            [name] =&gt; scripts
            [canonical] =&gt; script
            [see] =&gt; 
            [description] =&gt; 
        )

    [scrutability] =&gt; Array
        (
            [slug] =&gt; scrutability
            [parent] =&gt; 
            [name] =&gt; scrutability
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; 
        )

    [search] =&gt; Array
        (
            [slug] =&gt; search
            [parent] =&gt; 
            [name] =&gt; search
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; 
            [subentries] =&gt; Array
                (
                    [search!adversarial] =&gt; Array
                        (
                            [slug] =&gt; search!adversarial
                            [parent] =&gt; search
                            [name] =&gt; adversarial
                            [canonical] =&gt; 
                            [see] =&gt; 
                            [description] =&gt; 
                        )

                    [search!best first] =&gt; Array
                        (
                            [slug] =&gt; search!best first
                            [parent] =&gt; search
                            [name] =&gt; best first
                            [canonical] =&gt; 
                            [see] =&gt; best first search
                            [description] =&gt; 
                        )

                    [search!breadth first] =&gt; Array
                        (
                            [slug] =&gt; search!breadth first
                            [parent] =&gt; search
                            [name] =&gt; breadth first
                            [canonical] =&gt; 
                            [see] =&gt; breadth first search
                            [description] =&gt; 
                        )

                    [search!closed list] =&gt; Array
                        (
                            [slug] =&gt; search!closed list
                            [parent] =&gt; search
                            [name] =&gt; closed list
                            [canonical] =&gt; 
                            [see] =&gt; closed list
                            [description] =&gt; 
                        )

                    [search!depth first] =&gt; Array
                        (
                            [slug] =&gt; search!depth first
                            [parent] =&gt; search
                            [name] =&gt; depth first
                            [canonical] =&gt; 
                            [see] =&gt; depth first search
                            [description] =&gt; 
                        )

                    [search!deterministic] =&gt; Array
                        (
                            [slug] =&gt; search!deterministic
                            [parent] =&gt; search
                            [name] =&gt; deterministic
                            [canonical] =&gt; 
                            [see] =&gt; deterministic search
                            [description] =&gt; 
                        )

                    [search!feasibility] =&gt; Array
                        (
                            [slug] =&gt; search!feasibility
                            [parent] =&gt; search
                            [name] =&gt; feasibility
                            [canonical] =&gt; 
                            [see] =&gt; feasibility
                            [description] =&gt; 
                        )

                    [search!generate and test] =&gt; Array
                        (
                            [slug] =&gt; search!generate and test
                            [parent] =&gt; search
                            [name] =&gt; generate and test
                            [canonical] =&gt; 
                            [see] =&gt; generate and test
                            [description] =&gt; 
                        )

                    [search!graph] =&gt; Array
                        (
                            [slug] =&gt; search!graph
                            [parent] =&gt; search
                            [name] =&gt; graph
                            [canonical] =&gt; 
                            [see] =&gt; 
                            [description] =&gt; 
                        )

                    [search!heuristic] =&gt; Array
                        (
                            [slug] =&gt; search!heuristic
                            [parent] =&gt; search
                            [name] =&gt; heuristic
                            [canonical] =&gt; 
                            [see] =&gt; 
                            [description] =&gt; 
                        )

                    [search!open list] =&gt; Array
                        (
                            [slug] =&gt; search!open list
                            [parent] =&gt; search
                            [name] =&gt; open list
                            [canonical] =&gt; 
                            [see] =&gt; 
                            [description] =&gt; 
                        )

                    [search!optimality] =&gt; Array
                        (
                            [slug] =&gt; search!optimality
                            [parent] =&gt; search
                            [name] =&gt; optimality
                            [canonical] =&gt; 
                            [see] =&gt; 
                            [description] =&gt; 
                        )

                    [search!optimization] =&gt; Array
                        (
                            [slug] =&gt; search!optimization
                            [parent] =&gt; search
                            [name] =&gt; optimization
                            [canonical] =&gt; 
                            [see] =&gt; 
                            [description] =&gt; 
                        )

                    [search!satisficing] =&gt; Array
                        (
                            [slug] =&gt; search!satisficing
                            [parent] =&gt; search
                            [name] =&gt; satisficing
                            [canonical] =&gt; 
                            [see] =&gt; 
                            [description] =&gt; 
                        )

                    [search!solution path] =&gt; Array
                        (
                            [slug] =&gt; search!solution path
                            [parent] =&gt; search
                            [name] =&gt; solution path
                            [canonical] =&gt; 
                            [see] =&gt; 
                            [description] =&gt; 
                        )

                    [search!solution state] =&gt; Array
                        (
                            [slug] =&gt; search!solution state
                            [parent] =&gt; search
                            [name] =&gt; solution state
                            [canonical] =&gt; 
                            [see] =&gt; 
                            [description] =&gt; 
                        )

                )

        )

    [search engines] =&gt; Array
        (
            [slug] =&gt; search engines
            [parent] =&gt; 
            [name] =&gt; search engines
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; 
        )

    [search horizon] =&gt; Array
        (
            [slug] =&gt; search horizon
            [parent] =&gt; 
            [name] =&gt; search horizon
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; 
        )

    [search tree] =&gt; Array
        (
            [slug] =&gt; search tree
            [parent] =&gt; 
            [name] =&gt; search tree
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; 
            [subentries] =&gt; Array
                (
                    [search tree!depth] =&gt; Array
                        (
                            [slug] =&gt; search tree!depth
                            [parent] =&gt; search tree
                            [name] =&gt; depth
                            [canonical] =&gt; 
                            [see] =&gt; 
                            [description] =&gt; 
                        )

                )

        )

    [search trees] =&gt; Array
        (
            [slug] =&gt; search trees
            [parent] =&gt; 
            [name] =&gt; search trees
            [canonical] =&gt; search tree
            [see] =&gt; 
            [description] =&gt; 
        )

    [search tree!depth] =&gt; Array
        (
            [slug] =&gt; search tree!depth
            [parent] =&gt; search tree
            [name] =&gt; depth
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; 
        )

    [search!adversarial] =&gt; Array
        (
            [slug] =&gt; search!adversarial
            [parent] =&gt; search
            [name] =&gt; adversarial
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; 
        )

    [search!best first] =&gt; Array
        (
            [slug] =&gt; search!best first
            [parent] =&gt; search
            [name] =&gt; best first
            [canonical] =&gt; 
            [see] =&gt; best first search
            [description] =&gt; 
        )

    [search!breadth first] =&gt; Array
        (
            [slug] =&gt; search!breadth first
            [parent] =&gt; search
            [name] =&gt; breadth first
            [canonical] =&gt; 
            [see] =&gt; breadth first search
            [description] =&gt; 
        )

    [search!closed list] =&gt; Array
        (
            [slug] =&gt; search!closed list
            [parent] =&gt; search
            [name] =&gt; closed list
            [canonical] =&gt; 
            [see] =&gt; closed list
            [description] =&gt; 
        )

    [search!depth first] =&gt; Array
        (
            [slug] =&gt; search!depth first
            [parent] =&gt; search
            [name] =&gt; depth first
            [canonical] =&gt; 
            [see] =&gt; depth first search
            [description] =&gt; 
        )

    [search!deterministic] =&gt; Array
        (
            [slug] =&gt; search!deterministic
            [parent] =&gt; search
            [name] =&gt; deterministic
            [canonical] =&gt; 
            [see] =&gt; deterministic search
            [description] =&gt; 
        )

    [search!feasibility] =&gt; Array
        (
            [slug] =&gt; search!feasibility
            [parent] =&gt; search
            [name] =&gt; feasibility
            [canonical] =&gt; 
            [see] =&gt; feasibility
            [description] =&gt; 
        )

    [search!generate and test] =&gt; Array
        (
            [slug] =&gt; search!generate and test
            [parent] =&gt; search
            [name] =&gt; generate and test
            [canonical] =&gt; 
            [see] =&gt; generate and test
            [description] =&gt; 
        )

    [search!graph] =&gt; Array
        (
            [slug] =&gt; search!graph
            [parent] =&gt; search
            [name] =&gt; graph
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; 
        )

    [search!heuristic] =&gt; Array
        (
            [slug] =&gt; search!heuristic
            [parent] =&gt; search
            [name] =&gt; heuristic
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; 
        )

    [search!open list] =&gt; Array
        (
            [slug] =&gt; search!open list
            [parent] =&gt; search
            [name] =&gt; open list
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; 
        )

    [search!optimality] =&gt; Array
        (
            [slug] =&gt; search!optimality
            [parent] =&gt; search
            [name] =&gt; optimality
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; 
        )

    [search!optimization] =&gt; Array
        (
            [slug] =&gt; search!optimization
            [parent] =&gt; search
            [name] =&gt; optimization
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; 
        )

    [search!satisficing] =&gt; Array
        (
            [slug] =&gt; search!satisficing
            [parent] =&gt; search
            [name] =&gt; satisficing
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; 
        )

    [search!solution path] =&gt; Array
        (
            [slug] =&gt; search!solution path
            [parent] =&gt; search
            [name] =&gt; solution path
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; 
        )

    [search!solution state] =&gt; Array
        (
            [slug] =&gt; search!solution state
            [parent] =&gt; search
            [name] =&gt; solution state
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; 
        )

    [searle] =&gt; Array
        (
            [slug] =&gt; searle
            [parent] =&gt; 
            [name] =&gt; searle
            [canonical] =&gt; searle, john
            [see] =&gt; 
            [description] =&gt; 
        )

    [searle, j] =&gt; Array
        (
            [slug] =&gt; searle, j
            [parent] =&gt; 
            [name] =&gt; searle, j
            [canonical] =&gt; searle, john
            [see] =&gt; 
            [description] =&gt; 
        )

    [searle, john] =&gt; Array
        (
            [slug] =&gt; searle, john
            [parent] =&gt; 
            [name] =&gt; Searle, John
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; John Searle is a philosopher famous within AI for his {{Chinese room argument}}, which holds that true machine intelligence is not possible because there is no {{intentlonality}} in a machine.,, words mean nothing to it.
        )

    [seasonal adjustment] =&gt; Array
        (
            [slug] =&gt; seasonal adjustment
            [parent] =&gt; 
            [name] =&gt; seasonal adjustment
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; 
        )

    [seasonal adjustments] =&gt; Array
        (
            [slug] =&gt; seasonal adjustments
            [parent] =&gt; 
            [name] =&gt; seasonal adjustments
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; 
        )

    [segmentation] =&gt; Array
        (
            [slug] =&gt; segmentation
            [parent] =&gt; 
            [name] =&gt; segmentation
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; 
        )

    [self learning] =&gt; Array
        (
            [slug] =&gt; self learning
            [parent] =&gt; 
            [name] =&gt; self learning
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; Most {{machine learning}} is based on using large volumes of human-produced data.  This is even true of the way most {{unsupervised learning}} is used, as the data comes from previous human activity or data collection, such as medical records.  However, as models get ever larger, the available training data becomes a major bottleneck..   &lt;em&gt;Self-learning&lt;/em&gt; systems seek to go beyond the avaiable data in soem cases the AI system gathers its own data through experimentation, as in {{reinforcement learning}}, or where AI sysems are pitted against each other, as found in {{self play}} and {{adversarial learning}}.  These processes may be {{bootstrapped}} using initial human-created data (as in {{AlphaGo}} or work entirely from scratch using some form of specification of the problems (as in {{AlphaZero}}).
        )

    [self play] =&gt; Array
        (
            [slug] =&gt; self play
            [parent] =&gt; 
            [name] =&gt; self play
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; Self play is used in {{gamp playing}} AI in order to grow the available data of games.  Game playing systems can be trained using records of past human games, for example, extensive records of chess competitions,  However, this apprach is limited by the available data.  Systems such as {{AlphaGo}} and its successors grew their available data by pitting AI aganst AI, this being able to create vast numbers of entirely machine-generated games.  This data can then be used to train better game playing AI, and these used to generate yet more games.
        )

    [self-organising map] =&gt; Array
        (
            [slug] =&gt; self-organising map
            [parent] =&gt; 
            [name] =&gt; self-organising map
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; A self-organising map (SOM) is a form of {{unsupervised learning}} where nodes on a 2D grid (or more geometric lattice structure) act as archetypes or key values for a form of clustering, where like items are allocated to grid points that have some form of relationship.    The prime example of a SOM are {{Kohonen networks}}.  Like other forms of {{clustering algorithm}}, SOM can be used for data interpretation where a human expert labels areas on the map, or as a form of {[dimension reduction}} either using the nodes as discrete {{categorical}} clusters or using the 2D (or higer diennsional) embedding of the grid or lattice as {{continuous}} dimensions.
        )

    [self-organising maps] =&gt; Array
        (
            [slug] =&gt; self-organising maps
            [parent] =&gt; 
            [name] =&gt; self-organising maps
            [canonical] =&gt; self-organising map
            [see] =&gt; 
            [description] =&gt; 
        )

    [self-organizing map] =&gt; Array
        (
            [slug] =&gt; self-organizing map
            [parent] =&gt; 
            [name] =&gt; self-organizing map
            [canonical] =&gt; self-organising map
            [see] =&gt; 
            [description] =&gt; 
        )

    [self-organizing network] =&gt; Array
        (
            [slug] =&gt; self-organizing network
            [parent] =&gt; 
            [name] =&gt; self-organizing network
            [canonical] =&gt; 
            [see] =&gt; self-organising map
            [description] =&gt; 
        )

    [semantic analysis] =&gt; Array
        (
            [slug] =&gt; semantic analysis
            [parent] =&gt; 
            [name] =&gt; semantic analysis
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; 
        )

    [semantic desktop] =&gt; Array
        (
            [slug] =&gt; semantic desktop
            [parent] =&gt; 
            [name] =&gt; semantic desktop
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; 
        )

    [semantic grammar] =&gt; Array
        (
            [slug] =&gt; semantic grammar
            [parent] =&gt; 
            [name] =&gt; semantic grammar
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; 
        )

    [semantic integrity] =&gt; Array
        (
            [slug] =&gt; semantic integrity
            [parent] =&gt; 
            [name] =&gt; semantic integrity
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; 
        )

    [semantic network] =&gt; Array
        (
            [slug] =&gt; semantic network
            [parent] =&gt; 
            [name] =&gt; semantic network
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; 
        )

    [semantic networks] =&gt; Array
        (
            [slug] =&gt; semantic networks
            [parent] =&gt; 
            [name] =&gt; semantic networks
            [canonical] =&gt; semantic network
            [see] =&gt; 
            [description] =&gt; 
        )

    [semantic web] =&gt; Array
        (
            [slug] =&gt; semantic web
            [parent] =&gt; 
            [name] =&gt; semantic web
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; 
        )

    [semi-autonomous car] =&gt; Array
        (
            [slug] =&gt; semi-autonomous car
            [parent] =&gt; 
            [name] =&gt; semi-autonomous car
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; 
        )

    [semi-supervised learning] =&gt; Array
        (
            [slug] =&gt; semi-supervised learning
            [parent] =&gt; 
            [name] =&gt; semi-supervised learning
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; 
        )

    [sensation] =&gt; Array
        (
            [slug] =&gt; sensation
            [parent] =&gt; 
            [name] =&gt; sensation
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; 
        )

    [senses] =&gt; Array
        (
            [slug] =&gt; senses
            [parent] =&gt; 
            [name] =&gt; senses
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; 
        )

    [sensitivity] =&gt; Array
        (
            [slug] =&gt; sensitivity
            [parent] =&gt; 
            [name] =&gt; sensitivity
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; 
        )

    [sensor fusion] =&gt; Array
        (
            [slug] =&gt; sensor fusion
            [parent] =&gt; 
            [name] =&gt; sensor fusion
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; 
        )

    [sentence parsing] =&gt; Array
        (
            [slug] =&gt; sentence parsing
            [parent] =&gt; 
            [name] =&gt; sentence parsing
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; 
            [subentries] =&gt; Array
                (
                    [sentence parsing!bottom up] =&gt; Array
                        (
                            [slug] =&gt; sentence parsing!bottom up
                            [parent] =&gt; sentence parsing
                            [name] =&gt; bottom up
                            [canonical] =&gt; 
                            [see] =&gt; 
                            [description] =&gt; 
                        )

                    [sentence parsing!methods] =&gt; Array
                        (
                            [slug] =&gt; sentence parsing!methods
                            [parent] =&gt; sentence parsing
                            [name] =&gt; methods
                            [canonical] =&gt; 
                            [see] =&gt; 
                            [description] =&gt; 
                        )

                    [sentence parsing!systems] =&gt; Array
                        (
                            [slug] =&gt; sentence parsing!systems
                            [parent] =&gt; sentence parsing
                            [name] =&gt; systems
                            [canonical] =&gt; 
                            [see] =&gt; 
                            [description] =&gt; 
                        )

                    [sentence parsing!top down] =&gt; Array
                        (
                            [slug] =&gt; sentence parsing!top down
                            [parent] =&gt; sentence parsing
                            [name] =&gt; top down
                            [canonical] =&gt; 
                            [see] =&gt; 
                            [description] =&gt; 
                        )

                )

        )

    [sentence parsing!bottom up] =&gt; Array
        (
            [slug] =&gt; sentence parsing!bottom up
            [parent] =&gt; sentence parsing
            [name] =&gt; bottom up
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; 
        )

    [sentence parsing!methods] =&gt; Array
        (
            [slug] =&gt; sentence parsing!methods
            [parent] =&gt; sentence parsing
            [name] =&gt; methods
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; 
        )

    [sentence parsing!systems] =&gt; Array
        (
            [slug] =&gt; sentence parsing!systems
            [parent] =&gt; sentence parsing
            [name] =&gt; systems
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; 
        )

    [sentence parsing!top down] =&gt; Array
        (
            [slug] =&gt; sentence parsing!top down
            [parent] =&gt; sentence parsing
            [name] =&gt; top down
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; 
        )

    [sentence parsing methods] =&gt; Array
        (
            [slug] =&gt; sentence parsing methods
            [parent] =&gt; 
            [name] =&gt; sentence parsing methods
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; 
        )

    [sentence-level processing] =&gt; Array
        (
            [slug] =&gt; sentence-level processing
            [parent] =&gt; 
            [name] =&gt; sentence-level processing
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; 
        )

    [sequence order] =&gt; Array
        (
            [slug] =&gt; sequence order
            [parent] =&gt; 
            [name] =&gt; sequence order
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; 
        )

    [set theory] =&gt; Array
        (
            [slug] =&gt; set theory
            [parent] =&gt; 
            [name] =&gt; set theory
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; 
        )

    [sexual selection] =&gt; Array
        (
            [slug] =&gt; sexual selection
            [parent] =&gt; 
            [name] =&gt; sexual selection
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; 
        )

    [shard] =&gt; Array
        (
            [slug] =&gt; shard
            [parent] =&gt; 
            [name] =&gt; shard
            [canonical] =&gt; sharding
            [see] =&gt; 
            [description] =&gt; 
        )

    [sharding] =&gt; Array
        (
            [slug] =&gt; sharding
            [parent] =&gt; 
            [name] =&gt; sharding
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; Sharding is used in {{big data}} applications to divide the data into portions (called shards) so that they can be stored and processed semi-independently.  If we think of simple tabel data, the shards may be split along the rows, so that all the data for a single item (row) is always stored together.  Alternatively particular columns (item fields/features) may be stored sepaartely, linked by some form of {{key}} or {{identifier}}.  The nature of the sharding affects the efficiency of algorithms, so there is often a desire to put data that needs to be processed together in the same shard.  This is particularly difficult for {{graph data}}, such as {{social network data}}, as the links between items inevitably cut across shards.
        )

    [sheep dogs] =&gt; Array
        (
            [slug] =&gt; sheep dogs
            [parent] =&gt; 
            [name] =&gt; sheep dogs
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; 
        )

    [shelley, mary] =&gt; Array
        (
            [slug] =&gt; shelley, mary
            [parent] =&gt; 
            [name] =&gt; Shelley, Mary
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; Mary Shelley is the author of the story of {{Frankenstein}}'s monster.
        )

    [shrdlu] =&gt; Array
        (
            [slug] =&gt; shrdlu
            [parent] =&gt; 
            [name] =&gt; SHRDLU
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; 
        )

    [sigmoid] =&gt; Array
        (
            [slug] =&gt; sigmoid
            [parent] =&gt; 
            [name] =&gt; sigmoid
            [canonical] =&gt; sigmoid function
            [see] =&gt; 
            [description] =&gt; 
        )

    [sigmoid activation function] =&gt; Array
        (
            [slug] =&gt; sigmoid activation function
            [parent] =&gt; 
            [name] =&gt; sigmoid activation function
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; A sigmoid activation function is used in many {{neural networks}}.  the properties of a {{sigmoid function}} mean that it shares some of the {{non-lonearity}} of a {{threshold function}}, but is {{differentiable}} enabling learning through the {{backpropogation algorithm}}.
        )

    [sigmoid function] =&gt; Array
        (
            [slug] =&gt; sigmoid function
            [parent] =&gt; 
            [name] =&gt; sigmoid function
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; A sigmoid function is a generic term refering to S-like functions that have a step-like nature, but are smooth without the {{discontinuity}} of the {{threshold function}}.  This maes them more emnable to use in {{machine learning}} as, in general, {{continuity}} aids learning.
        )

    [sigmoid functions] =&gt; Array
        (
            [slug] =&gt; sigmoid functions
            [parent] =&gt; 
            [name] =&gt; sigmoid functions
            [canonical] =&gt; sigmoid function
            [see] =&gt; 
            [description] =&gt; 
        )

    [signal processing] =&gt; Array
        (
            [slug] =&gt; signal processing
            [parent] =&gt; 
            [name] =&gt; signal processing
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; 
        )

    [similarity] =&gt; Array
        (
            [slug] =&gt; similarity
            [parent] =&gt; 
            [name] =&gt; similarity
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; 
        )

    [similarity matrix] =&gt; Array
        (
            [slug] =&gt; similarity matrix
            [parent] =&gt; 
            [name] =&gt; similarity matrix
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; 
        )

    [simon, h] =&gt; Array
        (
            [slug] =&gt; simon, h
            [parent] =&gt; 
            [name] =&gt; simon, h
            [canonical] =&gt; simon, herbert
            [see] =&gt; 
            [description] =&gt; 
        )

    [simon, herbert] =&gt; Array
        (
            [slug] =&gt; simon, herbert
            [parent] =&gt; 
            [name] =&gt; Simon, Herbert
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; Herbert A. Simon was a political economist and AI pioneer who, amnstt other things, coned the terms {{satisficing}} and {{bounded rationality}}.  He also, with {{Alan Newell}}, proposed the {{physical symbol system hypothesis}}.
        )

    [simulated annealing] =&gt; Array
        (
            [slug] =&gt; simulated annealing
            [parent] =&gt; 
            [name] =&gt; simulated annealing
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; 
        )

    [simulated data] =&gt; Array
        (
            [slug] =&gt; simulated data
            [parent] =&gt; 
            [name] =&gt; simulated data
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; 
        )

    [single-shot learning] =&gt; Array
        (
            [slug] =&gt; single-shot learning
            [parent] =&gt; 
            [name] =&gt; single-shot learning
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; 
        )

    [singularity] =&gt; Array
        (
            [slug] =&gt; singularity
            [parent] =&gt; 
            [name] =&gt; singularity
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; 
        )

    [skynet project] =&gt; Array
        (
            [slug] =&gt; skynet project
            [parent] =&gt; 
            [name] =&gt; Skynet
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; Skynet is the name for a system of CCTV cameras, {{facial recogntion}} technlogy and {{biometric}} databases deployed in mainland China.  It is used for criminal investigations, but also for political purposes.  Skynet is of course also the name of the comouter system that destroys most of humankind in the Terminator films.
        )

    [smart dust] =&gt; Array
        (
            [slug] =&gt; smart dust
            [parent] =&gt; 
            [name] =&gt; smart dust
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; Smart dust refers to tiny sensing and processing units designed to be scattered in large numbers, typically for sensing purposes. Each is individually poor fidelity but collectively they can deliver important information. Smart dust usually has some form of network communication to enable collective processing, but this needs to be low bandwidt and range because of the mited power availability of the devices.  The main uses are miitary and environmental (although the smart device units themselves are polluting if not recovered).
        )

    [smoothing] =&gt; Array
        (
            [slug] =&gt; smoothing
            [parent] =&gt; 
            [name] =&gt; smoothing
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; 
        )

    [smoothing function] =&gt; Array
        (
            [slug] =&gt; smoothing function
            [parent] =&gt; 
            [name] =&gt; smoothing function
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; 
        )

    [snipit] =&gt; Array
        (
            [slug] =&gt; snipit
            [parent] =&gt; 
            [name] =&gt; Snip!t
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; 
        )

    [soar] =&gt; Array
        (
            [slug] =&gt; soar
            [parent] =&gt; 
            [name] =&gt; SOAR
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; SOAR is a {{cognitive architecture}} crated by {{John Laird}}, {{Alan Newell}} and {{Paul Rosenbloom}}.
        )

    [sobel's operator] =&gt; Array
        (
            [slug] =&gt; sobel's operator
            [parent] =&gt; 
            [name] =&gt; Sobel's operator
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; Sobel'soperator is used to assess whether there is a line at a particular location and to estmiate its direction.  It can be seen as the sum of two factors, one measuring the horizontal rate of change and the other the virtical rate of change. &lt;br&gt;&lt;br&gt; &amp;nbsp; &amp;nbsp; &amp;nbsp; H &amp;nbsp; = &amp;nbsp; (c+2f+i) - (a+2d+g)  &lt;br&gt;
&amp;nbsp; &amp;nbsp; &amp;nbsp; V &amp;nbsp; = &amp;nbsp; (g+2h+i) - (a+2b+c)  &lt;br&gt;
&amp;nbsp; &amp;nbsp; &amp;nbsp; G  &amp;nbsp; = &amp;nbsp; | H |  +  | V | &lt;br&gt; where ,a,b,c,d,e,f,g,h,i are the pixels in a 3x3 square (see diagram)
        )

    [social machine] =&gt; Array
        (
            [slug] =&gt; social machine
            [parent] =&gt; 
            [name] =&gt; social machine
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; Those working in {{web science}} ofter refer to the web as a 'social machine' as the human users and authors of the web through their social interactions captured in the web (e.g. through social media) and those mediated by it (e.g. publishing pages for others to view) can be regarded as collectively processing.
        )

    [social media] =&gt; Array
        (
            [slug] =&gt; social media
            [parent] =&gt; 
            [name] =&gt; social media
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; 
        )

    [software agents] =&gt; Array
        (
            [slug] =&gt; software agents
            [parent] =&gt; 
            [name] =&gt; software agents
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; 
        )

    [software ring-fence] =&gt; Array
        (
            [slug] =&gt; software ring-fence
            [parent] =&gt; 
            [name] =&gt; software ring-fence
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; 
        )

    [sparql] =&gt; Array
        (
            [slug] =&gt; sparql
            [parent] =&gt; 
            [name] =&gt; SPARQL
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; SPARQL is a query language for {{RDF}} data.  It is a form of {{graph query}} for triple-based representations.
        )

    [sparse matrices] =&gt; Array
        (
            [slug] =&gt; sparse matrices
            [parent] =&gt; 
            [name] =&gt; sparse matrices
            [canonical] =&gt; sparse matrix
            [see] =&gt; 
            [description] =&gt; 
        )

    [sparse matrix] =&gt; Array
        (
            [slug] =&gt; sparse matrix
            [parent] =&gt; 
            [name] =&gt; sparse matrix
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; A {{matrix}} is called sparse if most of its entries are zero (or empty).  Sparse matrices can be store efficiently by only recording the entroes with actual values.  However, this can make algorithms to deal with sparse matrices more complicated.
        )

    [spectrogram] =&gt; Array
        (
            [slug] =&gt; spectrogram
            [parent] =&gt; 
            [name] =&gt; spectrogram
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; A spectrogram shows a sound, such as speach, based on its frequency components. 
        )

    [speech acts] =&gt; Array
        (
            [slug] =&gt; speech acts
            [parent] =&gt; 
            [name] =&gt; speech acts
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; 
        )

    [speech recognition] =&gt; Array
        (
            [slug] =&gt; speech recognition
            [parent] =&gt; 
            [name] =&gt; speech recognition
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; 
        )

    [spikes] =&gt; Array
        (
            [slug] =&gt; spikes
            [parent] =&gt; 
            [name] =&gt; spikes
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; When {{biological neurones}} pass potential to one another, they do not do so in lock step and using continuous vvalues, as is usually the case in {[artificial neutal neteotks}}, but as short spikes.  These are more likely to occur when a cell is highly activated, leading to more frequent spikes, and the continuous value approxmates this, but the spiking cam lead to differet behaviours.  This more detail bilogval model is te bases if {{spiking neural networks}}.
        )

    [spiking neural networks] =&gt; Array
        (
            [slug] =&gt; spiking neural networks
            [parent] =&gt; 
            [name] =&gt; spiking neural networks
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; A spiking neural network emulates the way that real {{neurones}} in the brain communicate through discrete {{spikes}} rather than weighted links as is most common in {{artificial neural networks}}.  In some ways the two behave similarly and a weighted impulse is rougly equivalent to a fast firing spike link.  However, there are differences, for example the potential for {{disambiguation}}.
        )

    [spin glass models] =&gt; Array
        (
            [slug] =&gt; spin glass models
            [parent] =&gt; 
            [name] =&gt; spin glass models
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; Spin glass models are used to physics to explain the way that magnetic materials that initially have spins (generating their magnetis moment) in random directiosn, settle down to have spins aligned over patches or the entire material.  They are used as an analogy for {{subsymbolic}} {{machine learning}} including {{Bolzmann machines}} and {{Hopfield networks}}.
        )

    [sporadic] =&gt; Array
        (
            [slug] =&gt; sporadic
            [parent] =&gt; 
            [name] =&gt; sporadic
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; 
        )

    [spreading activation] =&gt; Array
        (
            [slug] =&gt; spreading activation
            [parent] =&gt; 
            [name] =&gt; spreading activation
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; 
        )

    [square matrices] =&gt; Array
        (
            [slug] =&gt; square matrices
            [parent] =&gt; 
            [name] =&gt; square matrices
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; A square matrix is a {{matrix}} with the same number of rows and columns.
        )

    [standard logic] =&gt; Array
        (
            [slug] =&gt; standard logic
            [parent] =&gt; 
            [name] =&gt; standard logic
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; Standard logic uses simple facts and rules abiut a single situation.  For example, the classic &quot;Socrates is a man; all men are mortal, therefore Socrates is mortal&quot;.   This example includes named  things (Socrates), {{predicates}} (properties of things, such as &quot;is a man&quot;), of {{propositions}} (facts such as &quot;Socrates is a man&quot;),  {{quantifiers}} (ways talking about lots of things at once, such as &quot;all&quot;) and {{syllogisms}} (reasoning).   These are often written in formal terms such as: &lt;br&gt; &amp;nbsp; &amp;nbsp; &amp;nbsp; is_a_man(Socrates)   &lt;br&gt; &amp;nbsp; &amp;nbsp; &amp;nbsp; &amp;forall; p :  is_a_man(p) &amp;Implies;   mortal(p)  &lt;br&gt; &amp;nbsp; &amp;nbsp; &amp;nbsp; mortal(Socrates)  &amp;nbsp; &amp;nbsp; -- from the rule  ( &amp;forall; x :  P(x) &amp; Implies;   Q(x)  ) and P(y) infer  Q(y)  &lt;br&gt; Note how there are two levels of implication here the implies in the propositions (&amp;Implies; ) and the rule implication.  The former that can be regarded as a {{syntactic sugar}}  for the equivaent ands and ors: &lt;br&gt; &amp;nbsp; &amp;nbsp; &amp;nbsp; X &amp;Implies; Y. &amp;equiv;  &amp;not; X &amp;or; Y  &lt;br&gt; &lt;br&gt; Standard logics  can be contrasted with {{non-standard logic}} such as {{non-monotonic reasoning}}, {{modal logic}} or {{fuzzy logic}}
        )

    [standard logics] =&gt; Array
        (
            [slug] =&gt; standard logics
            [parent] =&gt; 
            [name] =&gt; standard logics
            [canonical] =&gt; standard logic
            [see] =&gt; 
            [description] =&gt; 
        )

    [state space search] =&gt; Array
        (
            [slug] =&gt; state space search
            [parent] =&gt; 
            [name] =&gt; state space search
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; 
        )

    [state space traversal] =&gt; Array
        (
            [slug] =&gt; state space traversal
            [parent] =&gt; 
            [name] =&gt; state space traversal
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; 
        )

    [statistical techniques] =&gt; Array
        (
            [slug] =&gt; statistical techniques
            [parent] =&gt; 
            [name] =&gt; statistical techniques
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; 
        )

    [status change event] =&gt; Array
        (
            [slug] =&gt; status change event
            [parent] =&gt; 
            [name] =&gt; status change event
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; A status change event is an {{event}} that is registered when some sensor or digital value changes.  For example, a change in the temeprature in a central heating systems, a finger movement in a {{user interface}}, or a share price .on a stock exchange.  The event may be used to {{trigger}} an action in a software {{agent}} or rase a {{notification}} for a user.  Discrete changes in status, such as an in/out indicator on an office door may have an idetifiable event that causes the change, and thus this can be monitored.  However sometime other means such as periodic {{polling}} are necessary.  For continuous values often very small changes are ignored and the status change event only raised wen the change exceeds some {{threshold}}.
        )

    [stemming] =&gt; Array
        (
            [slug] =&gt; stemming
            [parent] =&gt; 
            [name] =&gt; stemming
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; Stemming is a process of reducing a word to its stem as a normalised form for the purposes of {{natural language processing}}, {{indexing}} or {{document retreval}}.  For example, plurals are reduced to their singular form by deleting the trailing 's' and gerunds of verbs have theor 'ing' removed.  Stemming is usually perfomed at a {{lexical}} level without any {{semantic knowledge}}, which can lead to a small proportion of inappropriate stems, for example,  'starling' might be stemmed to 'starl'.
        )

    [stereo vision] =&gt; Array
        (
            [slug] =&gt; stereo vision
            [parent] =&gt; 
            [name] =&gt; stereo vision
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; Stereo vision uses two (or more) cameras in order, inter alia, to be able to detect the distance of objects using {{triangulation}}.
        )

    [stimulus-response] =&gt; Array
        (
            [slug] =&gt; stimulus-response
            [parent] =&gt; 
            [name] =&gt; stimulus-response
            [canonical] =&gt; stimulus-response learning
            [see] =&gt; 
            [description] =&gt; 
        )

    [stimulus-response learning] =&gt; Array
        (
            [slug] =&gt; stimulus-response learning
            [parent] =&gt; 
            [name] =&gt; stimulus-response learning
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; 
        )

    [stop words] =&gt; Array
        (
            [slug] =&gt; stop words
            [parent] =&gt; 
            [name] =&gt; stop words
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; In {{natural language processing}} stop words are words that are removed or ignored before further processing.  They are usually the most common words in language such as 'the', 'a', or 'is'.
        )

    [strong ai] =&gt; Array
        (
            [slug] =&gt; strong ai
            [parent] =&gt; 
            [name] =&gt; strong AI
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; Strong AI is the belief that computers can (and will) be able to achieve full {{human-like intelligence}}, in part because human intelligence itself, so it would be claimed, is effectively built on a biological computer, the brain.  Strong AI stands in contrast to {{weak AI}} which only seeks computer systems that perform in an intelligent manner even if they do so in very different ways to humans.
        )

    [strong learner] =&gt; Array
        (
            [slug] =&gt; strong learner
            [parent] =&gt; 
            [name] =&gt; strong learner
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; In {{machine learner}} a strong learner is a model that on ist own is suffciently effective to be used for all data.  It is in contrast to a {{weak learner}} that in some way would need to be combined with ither weak learners to be useful.
        )

    [stuttering in search trees] =&gt; Array
        (
            [slug] =&gt; stuttering in search trees
            [parent] =&gt; 
            [name] =&gt; stuttering in search trees
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; Stuttering in search trees arises when moves in the {{search tree}} can go back and forth between the same states.  For example, in the {{Towers of Hanoi}} one could repeatedly move a ring form tower 1 to 2 on the forts movem then form 2 to one on the next move and so on.  A {{closed list}} can be used to record visited states and hence avoid stuttering.
        )

    [stuxnet] =&gt; Array
        (
            [slug] =&gt; stuxnet
            [parent] =&gt; 
            [name] =&gt; Stuxnet
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; Stuxnet was the worlds first known true {{cyber-weapon}}.   Stuxnet was a {{worm}} transported on USB pen-drive sticks and designed to target centrifuges at Iranian nuclear processing facilities.  Due to its complexity, it was widely believed to have been a joint effort of security services in Israel and USA, who wished to prevent Iran obtaining nuclear material for weapons, but neither ever confirmed this.  It targetted parricular micro-processors that were used in the centrifuges and instructed them to drive the centrifuge motors in patterns that physically damaged them.  However, the same micro-processor was used in other kinds of equipment and there were unconformed reports that Stuxnet had been found in Ukrainian and Russian nuclear power stations.
        )

    [sub-symbolic] =&gt; Array
        (
            [slug] =&gt; sub-symbolic
            [parent] =&gt; 
            [name] =&gt; sub-symbolic
            [canonical] =&gt; sub-symbolic systems
            [see] =&gt; 
            [description] =&gt; 
        )

    [sub-symbolic systems] =&gt; Array
        (
            [slug] =&gt; sub-symbolic systems
            [parent] =&gt; 
            [name] =&gt; sub-symbolic systems
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; Subsymbolic systems are those which use low-level representations such as the weights in a {{neural newtworks}} or the artificial insects in {{swarm computing}}.  One could argue that there is still digital symbol manipulation when they execute, but not at a recognisable level.  In contrast {{symbolic systems}} use symbols such as names, places, and concepts whch are manipulated by explicit rules.
        )

    [subject-predicate-object triples] =&gt; Array
        (
            [slug] =&gt; subject-predicate-object triples
            [parent] =&gt; 
            [name] =&gt; subject--predicate--object triples
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; Subject-predicate-object triples (or simply {{triples}}) are the main form of {{knowledge representation}} in {{RDF}}, the language of the {{semantic web}}.   Becasue of this {{RDF databases}} are often referd to has {{triple stores}}.  Although the {{triple}} seems verey simple, it can dscribe anything that more complex representations such as {{tables}}, {{XML}} or {{JSON data structures}}   Some {{RDF databases}} also stupport {{named graphs}}, where {{triples}} are tagged with a graph URI, and hence are called {{quad stores}}.&lt;br&gt;&lt;br&gt;An example: &lt;br&gt; &lt;table align=&quot;center&quot;&gt;&lt;tr&gt;&lt;th align=&quot;center&quot;&gt;subject&lt;/th&gt;&lt;td&gt;&lt;/td&gt;&lt;th align=&quot;center&quot;&gt;predicate&lt;/th&gt;&lt;td&gt;&lt;/td&gt;&lt;th align=&quot;center&quot;&gt;object&lt;/th&gt;&lt;/tr&gt;&lt;tr&gt;&lt;td align=&quot;center&quot;&gt;Scooby Doo&lt;/td&gt;&lt;td align=&quot;center&quot;&gt;&ndash;-&lt;/td&gt;&lt;td align=&quot;center&quot;&gt;instance of&lt;/td&gt;&lt;td align=&quot;center&quot;&gt;&ndash;-&lt;/td&gt;&lt;td align=&quot;center&quot;&gt;Great Dane&lt;/td&gt;&lt;/tr&gt;&lt;tr&gt;&lt;td align=&quot;center&quot;&gt;Dog&lt;/td&gt;&lt;td align=&quot;center&quot;&gt;&ndash;-&lt;/td&gt;&lt;td align=&quot;center&quot;&gt;number of legs&lt;/td&gt;&lt;td align=&quot;center&quot;&gt;&ndash;-&lt;/td&gt;&lt;td align=&quot;center&quot;&gt;4&lt;/td&gt;&lt;/tr&gt;&lt;/table&gt;&lt;br&gt;In {{RDF}} the subject and predicate must both be {{URI}}s, like this: &lt;br&gt; &lt;table align=&quot;center&quot;&gt;&lt;tr&gt;&lt;td align=&quot;center&quot;&gt;&amp;lt;https://alandix.com/&amp;gt;&lt;/td&gt;&lt;td align=&quot;center&quot;&gt;&amp;lt;dc:creator&amp;gt;&lt;/td&gt;           &lt;td align=&quot;center&quot;&gt;&quot;Alan Dix&quot;&lt;/td&gt;&lt;/tr&gt;
&lt;tr&gt;&lt;td align=&quot;center&quot;&gt;&amp;lt;aibk:Scooby_Doo&amp;gt;&lt;/td&gt;       &lt;td align=&quot;center&quot;&gt;&amp;lt;rdf:type&amp;gt;&lt;/td&gt;             &lt;td align=&quot;center&quot;&gt;&amp;lt;aibk:Great_Dane&amp;gt;&lt;/td&gt;&lt;/tr&gt;
&lt;tr&gt;&lt;td align=&quot;center&quot;&gt;&amp;lt;aibk:Dog&amp;gt;&lt;/td&gt;              &lt;td align=&quot;center&quot;&gt;&amp;lt;aibk:number_of_legs&amp;gt;&lt;/td&gt;  &lt;td align=&quot;center&quot;&gt;&quot;4&quot;^^xsd:integer&lt;/td&gt;&lt;/tr&gt;&lt;/table&gt;


&ndash; &ndash; 4&lt;/table&gt;
        )

    [subject&mdash;predicate&mdash;object triples] =&gt; Array
        (
            [slug] =&gt; subject&mdash;predicate&mdash;object triples
            [parent] =&gt; 
            [name] =&gt; subject&mdash;predicate&mdash;object triples
            [canonical] =&gt; subject-predicate-object triples
            [see] =&gt; 
            [description] =&gt; 
        )

    [successive refinement] =&gt; Array
        (
            [slug] =&gt; successive refinement
            [parent] =&gt; 
            [name] =&gt; successive refinement
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; 
        )

    [sufficient reason] =&gt; Array
        (
            [slug] =&gt; sufficient reason
            [parent] =&gt; 
            [name] =&gt; sufficient reason
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; In human communication we only feel the need to give sufficient reason for our xdecisions or actions, not unpack the firing of {{neorones}} in our brains.  This principle can be applied in {{explainable AI}}, when it is tempting to try to explain the minutae of, say, every node and oink in a {{deep neural network}}.   Instead, we should seek explanations at an appropriate level for the task at hand.
        )

    [supervised] =&gt; Array
        (
            [slug] =&gt; supervised
            [parent] =&gt; 
            [name] =&gt; supervised
            [canonical] =&gt; supervised learning
            [see] =&gt; 
            [description] =&gt; 
        )

    [supervised learning] =&gt; Array
        (
            [slug] =&gt; supervised learning
            [parent] =&gt; 
            [name] =&gt; supervised learning
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; In supervised learning the {{training set}} is {{labelled}}, that is it has some form of markers to indicate a desired grouping or value.  The supervised algorithm typically attempts to emulate or approximate the labelling of the training set.
        )

    [support vector machine] =&gt; Array
        (
            [slug] =&gt; support vector machine
            [parent] =&gt; 
            [name] =&gt; support vector machine
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; A support vector machine (SVM) is a {{multi-layer neural network}} where the first layer uses a large number of specialised {{non-linear functions}} in order to allow simpler leaerning in later levels.   Often the non-linear functions form a family, called a {{kernal}}, for example they maybe foms of {{wavelet}}.  Alongside other techniques such as {{reservoir ocmputing}} it is a way of creating {{non-linear diversity}}.  
        )

    [support vector machines] =&gt; Array
        (
            [slug] =&gt; support vector machines
            [parent] =&gt; 
            [name] =&gt; support vector machines
            [canonical] =&gt; support vector machine
            [see] =&gt; 
            [description] =&gt; 
        )

    [surrogate expert] =&gt; Array
        (
            [slug] =&gt; surrogate expert
            [parent] =&gt; 
            [name] =&gt; surrogate expert
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; 
        )

    [svm] =&gt; Array
        (
            [slug] =&gt; svm
            [parent] =&gt; 
            [name] =&gt; svm
            [canonical] =&gt; 
            [see] =&gt; support vector machine
            [description] =&gt; 
        )

    [swarm computing] =&gt; Array
        (
            [slug] =&gt; swarm computing
            [parent] =&gt; 
            [name] =&gt; swarm computing
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; Swarm computing is based on analogies with social insects such as ants and bees.  These insects behave in intelligent ways as a collective even though each individual insect is very simple.  In swarm computing many simple individual agents are prodused which explore a spece, for example to find an optmal solution.  They operate semi-independently with a level of simple communication that nudges their behaviour.  For examaple, if an agent in the swarm has found a good solution it may retrace the intermediate states it took to get there labelling them as good, but with decaying strengh, thus mirroring phermone trails.  Swarm computing is similar to {[ensemble method}}, but the latter are more about forming a diverse group of simple {[decsions making}} agents that together are used repeatedly to solve similar problems, wheres in swarm computing the agents are near identical and it usualy only the final result that matters.
        )

    [syllogism] =&gt; Array
        (
            [slug] =&gt; syllogism
            [parent] =&gt; 
            [name] =&gt; syllogism
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; A syllogism is a formal rule to combine logical premises. The classic is {{Aristotle}}'s: &lt;br&gt; &amp;nbsp; FROM &quot;all men are mortal &lt;br&gt; &amp;nbsp; AND &quot;Socrates is a man&quot; &lt;br&gt; &amp;nbsp; CONCLUDE: &quot;Socrates is mortal&quot;. &lt;br&gt;. This is using a general syllogism of the form: &lt;br&gt; &amp;nbsp; &amp;nbsp; FROM &quot;all members of CLASS(x) HAS_PROPERTY(y) &lt;br&gt; &amp;nbsp; AND &quot;ENTITY(z) is of CLASS(x)&quot; &lt;br&gt; &amp;nbsp; CONCLUDE: &quot;ENTITY(z)  HAS_PROPERTY(y)&quot;
        )

    [syllogisms] =&gt; Array
        (
            [slug] =&gt; syllogisms
            [parent] =&gt; 
            [name] =&gt; syllogisms
            [canonical] =&gt; syllogism
            [see] =&gt; 
            [description] =&gt; 
        )

    [symbolic regression] =&gt; Array
        (
            [slug] =&gt; symbolic regression
            [parent] =&gt; 
            [name] =&gt; symbolic regression
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; Symbolic regression is a technique in machine learning that attempts to create succinct equations to describe data.  It can be used in physics or engineering to suggest potential physical laws from empirical data.  Symbolic regression is named to suggest the connection with {{linear regression}}.  Whilst {{linear regression}} can only create linear relationships such as y = 3x+5, symbolic regression can look for complex formulae such as F = q(E&lt;sub&gt;f&lt;/sub&gt;  + B v sin &amp;theta;).
        )

    [symbolic systems] =&gt; Array
        (
            [slug] =&gt; symbolic systems
            [parent] =&gt; 
            [name] =&gt; symbolic systems
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; Symbolic systems use symbols such as names, places, and concepts whch are manipulated by explicit rules.  In contrast {{subsymbolic systems}}, such as {{neural networks}} use low-level representations such as the weights between nodes.
        )

    [symmetric matrices] =&gt; Array
        (
            [slug] =&gt; symmetric matrices
            [parent] =&gt; 
            [name] =&gt; symmetric matrices
            [canonical] =&gt; symmetric matrix
            [see] =&gt; 
            [description] =&gt; 
        )

    [symmetric matrix] =&gt; Array
        (
            [slug] =&gt; symmetric matrix
            [parent] =&gt; 
            [name] =&gt; symmetric matrix
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; A symmetric matrix is one where the elements either side of the diagonal morror oen another.  That is M&lt;sub &gt;i,j&lt;/sub&gt; = M&lt;sub &gt;j,i&lt;/sub&gt;, or, in other words, M is the matrix  transpose of itself. &lt;br&gt;A 3 x 3 symmetric matrix: &lt;br&gt; &lt;table align=&quot;center&quot;&gt;  &lt;tr&gt;&lt;td align=&quot;center&quot;&gt;3&lt;/td&gt;&lt;td align=&quot;center&quot;&gt;-1&lt;/td&gt;&lt;td align=&quot;center&quot;&gt;2&lt;/td&gt;&lt;/tr&gt;   &lt;tr&gt;&lt;td align=&quot;center&quot;&gt;-1&lt;/td&gt;&lt;td align=&quot;center&quot;&gt;7&lt;/td&gt;&lt;td align=&quot;center&quot;&gt;1&lt;/td&gt;&lt;/tr&gt;   &lt;tr&gt;&lt;td align=&quot;center&quot;&gt;2&lt;/td&gt;&lt;td align=&quot;center&quot;&gt;1&lt;/td&gt;&lt;td align=&quot;center&quot;&gt;-3&lt;/td&gt;&lt;/tr&gt; &lt;/table&gt;
        )

    [synapse weights] =&gt; Array
        (
            [slug] =&gt; synapse weights
            [parent] =&gt; 
            [name] =&gt; synapse weights
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; In the brain {{neurones}} connect at {{synapses}}.  This is where signals pass form one to another, but not all connectson are equally strong, the idea of synapse weights cacpures this strength.  In an {{artificial neural network}} these are encoded as a numerical value, but in the brain ithe synapses can physically grow or shrink depending on use, a key part of {{long-term memory}}.
        )

    [syntactic analysis] =&gt; Array
        (
            [slug] =&gt; syntactic analysis
            [parent] =&gt; 
            [name] =&gt; syntactic analysis
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; Syntactic analysis is the part of {{natural languag eprocessing}} that deals with the structure of the
sentence, that is its {{grammar}}.
        )

    [syntactic grammar] =&gt; Array
        (
            [slug] =&gt; syntactic grammar
            [parent] =&gt; 
            [name] =&gt; syntactic grammar
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; Syntactic grammars are based solely on the structure of words in sentences, but do not use any knowledge of the meaning of the sentance beyond {{part of speach}} or similar tags, such as whether a wird is a noun or verb.  Syntactic grammars are used in the {{syntactic analysis}} stage of {{natural language processing}} and the output is typically a {{parse tree}}.
        )

    [syntactic grammars] =&gt; Array
        (
            [slug] =&gt; syntactic grammars
            [parent] =&gt; 
            [name] =&gt; syntactic grammars
            [canonical] =&gt; syntactic grammar
            [see] =&gt; 
            [description] =&gt; 
        )

    [synthetic data] =&gt; Array
        (
            [slug] =&gt; synthetic data
            [parent] =&gt; 
            [name] =&gt; synthetic data
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; Synthetic data is data generated by models or my manipulating real data.  It can be useful for training {{deep neural networks}} when there is insufficient real data or real data is hard to obtain.  For example, 3D simulations can be used to emulate near accident situations, such as pedestraians crossing buy roads or in the dark; these will hopefully be rare in real gatgered video data, but invaluable to help {{autonomous cars}} respond well to exerme events. Synthetic data can also help generalisation, for example by adding noise to images or addings cropped, rotated or resized copies.  Synthetic data has also be sggestd as a way to reduce {{bias}} in data sets.
        )

    [system 1] =&gt; Array
        (
            [slug] =&gt; system 1
            [parent] =&gt; 
            [name] =&gt; System 1
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; In Daniel Kahneman's &quot;Thinking, Fast and Slow&quot;, System 1 refers to the rapid unconscious processes that allow instant reactions, for example, if you start to slip, or you reacxh to catch an approaching ball.  It is on contrats to System 2, which is slower more reasoned thinking.  System 1 is rapid, but may be hard to explain to others. In AI, {{neural networks}} an d{{subsymbolic AI}} in general correspond closely to human System 1 thinking.
        )

    [system 2] =&gt; Array
        (
            [slug] =&gt; system 2
            [parent] =&gt; 
            [name] =&gt; System 2
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; In Daniel Kahneman's &quot;Thinking, Fast and Slow&quot;, System 2 refers to reasoned, conscious thought process, for example weighingnup options for a holiday or deciding what kind of ball to pitch in a game before actually throwing it.  It is on contrast to System 1, which is the faster more unconscious reactions.  System 2 is relatively slow, but open to explanation, communication and discussion. In AI, traditional {{symbolic AI}} in corresponds most closely to human System 2 thinking.
        )

    [system event] =&gt; Array
        (
            [slug] =&gt; system event
            [parent] =&gt; 
            [name] =&gt; system event
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; A system event is when the {{software agent}} initiates some form of action based on an internal change, such as completing a computation, or a message from another {{agent}}.  This is in contrast to a {{user event}}, {{timed event}} or {{status change event}}.   In a {{user interface}} the term system event may refer to any event or action generated by the compter system for example an {{agent}} replying to a message or an {{expert system}} requesting information from the user. 
        )

    [system events] =&gt; Array
        (
            [slug] =&gt; system events
            [parent] =&gt; 
            [name] =&gt; system events
            [canonical] =&gt; system event
            [see] =&gt; 
            [description] =&gt; 
        )

    [systems response] =&gt; Array
        (
            [slug] =&gt; systems response
            [parent] =&gt; 
            [name] =&gt; systems response
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; The 'systems response' is a counter-argument to {{Searle's Chinese Room}}.  It argues that while the person in the room does not understand Chinese, the room as &lt;em&gt;system&lt;/em&gt; does.
        )

    [tacit knowledge] =&gt; Array
        (
            [slug] =&gt; tacit knowledge
            [parent] =&gt; 
            [name] =&gt; tacit knowledge
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; Tacit knowledge refers to things we know how to do or say, but don;t know that we know.  For example, you might walk the same route home everyday, but find it hard to describe to a friend how to go there.  This is important in {{expert systems}} as much of {{expert knowledge}} is tacit, so {{knowledge elicitation}} techniques need to {{externalise}} this.
        )

    [technological determinism] =&gt; Array
        (
            [slug] =&gt; technological determinism
            [parent] =&gt; 
            [name] =&gt; technological determinism
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; Technological determinism is the belief or attitude that technological developments are inevitable and so there is no point in challenging or fighting this 'progress'.  While there are certainly many clear examples where technological devlopments have transformed society, also examples where it is clear that the way in which this acts depends on the social, political, economic and geographical context of societies.  For example, there are numerous examples of wheeled toys, and indeed other wheels, in pre-Columbian Central America, but the wheel was not used for transport.
        )

    [template matching] =&gt; Array
        (
            [slug] =&gt; template matching
            [parent] =&gt; 
            [name] =&gt; template matching
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; Template matching is when text or data is matched aganst simple patterns (templates).  This can be appllied in {{natural language processing}} as an alternatuve to full {{parsing}}, especially in chatbots, for example, using a template such as &quot;tell me about X&quot; (using caps for variables).  It can also be used in {{structured data}}, for example matching &quot;&lt;p class=content&quot;&gt;CONTENT&lt;/p&gt;&quot; in a web page, and in {{data detectors}}.  In {{computer vision}} more compelx templates may be used to represent the visual appearance of objects to be matched against parts of an image.
        )

    [tensorflow] =&gt; Array
        (
            [slug] =&gt; tensorflow
            [parent] =&gt; 
            [name] =&gt; TensorFlow 
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; TensorFlow  is an open. source machine learning and library developed by Google originally for internal use and later released publically.
        )

    [terminal symbols] =&gt; Array
        (
            [slug] =&gt; terminal symbols
            [parent] =&gt; 
            [name] =&gt; terminal symbols
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; In a {{grammar}}, a terminal symbol is one that cannot be expanded into further symbols, for example a pronoun (such as 'it' or 'they').  In terms of a {{parse tree}}, {{non-terminal symbols}} usually label nodes wirthin the body of the tree, whereas terminal symbols label the leaves.
        )

    [text prediction] =&gt; Array
        (
            [slug] =&gt; text prediction
            [parent] =&gt; 
            [name] =&gt; text prediction
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; Text prediction is used to fill in possible following words of phrases given past text.  Simple versions based on {{n-gram}} frequencies can be used to offer typing suggestions or correct misspellings, but more complex predictions, especially those in {{large-language models}} can continue the text, almost indefinitely as found in chatbots such as {{ChatGPT}} and {{Bard}}.
        )

    [texture] =&gt; Array
        (
            [slug] =&gt; texture
            [parent] =&gt; 
            [name] =&gt; texture
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; In {{computer vision}} and {{graphics}} texture is fine-scale detail such as grass, skin or fabric weave, as opposed to areas of solid colour.  Some form of low-level processing is usually needed to detect that an area of texture is in some way all similar.  Ths may be performed using some form of statistical model or in the early layers of a {{deep neural network}}.
        )

    [theory of mind] =&gt; Array
        (
            [slug] =&gt; theory of mind
            [parent] =&gt; 
            [name] =&gt; theory of mind
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; Theory of mind refers to the ability to 'put oneself in anothers shoes' that is imagining the world from their view point. 
        )

    [thought experiment] =&gt; Array
        (
            [slug] =&gt; thought experiment
            [parent] =&gt; 
            [name] =&gt; thought experiment
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; A thought experiemnt is used in {{philosophy}} to help focus ona. problem or issue.  For example, {{Searle's Chinese room}} and {{Zombie Alice}} are used to discuss isues of {{consciousnes}} and the {{trolley prolem}} to explore ethical choices. 
        )

    [three-dimensional objects] =&gt; Array
        (
            [slug] =&gt; three-dimensional objects
            [parent] =&gt; 
            [name] =&gt; three-dimensional objects
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; In {{computer vision}} special algorithms and processes are needed to detect and identify three-dimensional objects from (usually) 2D images.  {{Waltz's algorithm}} is a classic image analysis method for single images following {{edge detection}}.
        )

    [threshold] =&gt; Array
        (
            [slug] =&gt; threshold
            [parent] =&gt; 
            [name] =&gt; threshold
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; A threshold is a critical value beyond which some event occurs or values change drastically.  In biological {{neorones}} and {{artificial neurons}} there is often no reaction for small or moderate amounts of activation from other neorons.  However, if the level of input activation goes beyond some threshold, then the neuron will fully fire.
        )

    [threshold function] =&gt; Array
        (
            [slug] =&gt; threshold function
            [parent] =&gt; 
            [name] =&gt; threshold function
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; A threshold function changes its value sharply when its input reaches a particular value (the threshold).  The simplest is a {{step function}} zero (or &ndash;1) when it input is less than zero and 1 thereafter.  A hard threshold is simple to implement, but can be hard the learn; so {{neural networks}} typcally use a {{non-linear}} {{sigmod function}} instead.
        )

    [thresholding] =&gt; Array
        (
            [slug] =&gt; thresholding
            [parent] =&gt; 
            [name] =&gt; thresholding
            [canonical] =&gt; threshold
            [see] =&gt; 
            [description] =&gt; 
        )

    [thresholding function] =&gt; Array
        (
            [slug] =&gt; thresholding function
            [parent] =&gt; 
            [name] =&gt; thresholding function
            [canonical] =&gt; threshold function
            [see] =&gt; 
            [description] =&gt; 
        )

    [tim berners-lee] =&gt; Array
        (
            [slug] =&gt; tim berners-lee
            [parent] =&gt; 
            [name] =&gt; Tim Berners-Lee
            [canonical] =&gt; 
            [see] =&gt; berners-lee, tim 
            [description] =&gt; 
        )

    [time dependence] =&gt; Array
        (
            [slug] =&gt; time dependence
            [parent] =&gt; 
            [name] =&gt; time dependence
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; A system is time dependent, if its behaviour changes over time based on its past history as well as its current external state.  For example, a simple electrical circuit with resistors will always give the same output for the same inputs ({{time independent}} behaviour), but if it includes capacitors it can have oscilatiry behaviour that depends on past inputs (time dependence).  Time dependence is usually are due to some form of hidden state (such as the levels of charge on capacitors in the electrical circuit) or unobserved state.  Many real phenomena, such as human psychology or the stock market exhibit time dependence, for example, if soemone plays a parctical joke your response may well depend on whether you are happy or stressed.  Materials wtih time dependent behaviour, such as collections of nanoparticles or biofilms, can be used in {{reservoir computing}} to allow it to model complex time dependent phenomena.
        )

    [time domain] =&gt; Array
        (
            [slug] =&gt; time domain
            [parent] =&gt; 
            [name] =&gt; time domain
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; When considering data that has a {{time series}}, such as sound level or weather patterns, we can represent it usingits value at successive time points (val(t=0), val(t=1), ...).  This is called a {{time domain}} representation, and is bot the simplest representation and also the form that most daat is initially gathered.  However, gthe time series can be represented in other ways, notabley {{frequency domain}} that treats the data as a sum of {{sine waves}} or similar functions.
        )

    [time domain data] =&gt; Array
        (
            [slug] =&gt; time domain data
            [parent] =&gt; 
            [name] =&gt; time domain data
            [canonical] =&gt; time domain
            [see] =&gt; 
            [description] =&gt; 
        )

    [timed event] =&gt; Array
        (
            [slug] =&gt; timed event
            [parent] =&gt; 
            [name] =&gt; timed event
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; 
        )

    [timed events] =&gt; Array
        (
            [slug] =&gt; timed events
            [parent] =&gt; 
            [name] =&gt; timed events
            [canonical] =&gt; timed event
            [see] =&gt; 
            [description] =&gt; A timed event is an {{event}} that occurs at a. specific time, typically {{triggering}} some behavior in a {{software agent}} or {{robot}}.  The timing may be fxef (e.g. at 4pm), periodic (e.g. every 5 second) or after a delay (e.g. after 100ms).  This is contrast to {{user events}}, {{system events}} or  {{status change events}} in the enviornment.  However, a perodic event is oftern used for {{polling}}, where a sensor is monitored in order to detect {{changes in status}}.
        )

    [tms] =&gt; Array
        (
            [slug] =&gt; tms
            [parent] =&gt; 
            [name] =&gt; TMS
            [canonical] =&gt; 
            [see] =&gt; truth maintenance system
            [description] =&gt; 
        )

    [top-down reasoning] =&gt; Array
        (
            [slug] =&gt; top-down reasoning
            [parent] =&gt; 
            [name] =&gt; top-down reasoning
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; 
        )

    [towers of hanoi] =&gt; Array
        (
            [slug] =&gt; towers of hanoi
            [parent] =&gt; 
            [name] =&gt; Towers of Hanoi
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; The Towers of Hanoi is a puzzle where the goal is to move rings between three towesr where smaller rings always have to lie upon larger ones.  
        )

    [transfer learning] =&gt; Array
        (
            [slug] =&gt; transfer learning
            [parent] =&gt; 
            [name] =&gt; transfer learning
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; Transfer learning is when a {{machine learning}} system that has been trained on one kind of data, can also be modified or easily retrained to operate on other kinds of data.  For example, a {{deep neural network}} trained on English text may learn deep grammatical patrerns as well as the precise words, and so be able to be modified, with less re-training, to operate on other Germanic languages.  Similarly, game playing systems may be abe to quickly pick up similar games as they have learnt aspects of deep strategy.
        )

    [transformational analogy] =&gt; Array
        (
            [slug] =&gt; transformational analogy
            [parent] =&gt; 
            [name] =&gt; transformational analogy
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; A transformational analogy is a form of {{reasoning}} where one finds a situation that is an {{analogy}} of the current problem, andf then attemtps to transform the solution of the analogus problem to find a sokutikn. to the current one.  For example, when faced with thenproblem of smplifying x&lt;sup&gt;3&lt;/sup&gt; &amp;times; x&lt;sup&gt;5&lt;/sup&gt;, one might use an analogy with 3x + 5x which you know simplfes ot 8x.  Using the analogy that  in algebra (multiplication.power) behaves like (adding,multiplying by numbers), you might then reason that  x&lt;sup&gt;3&lt;/sup&gt; &amp;times; x&lt;sup&gt;5&lt;/sup&gt; = x&lt;sup&gt;8&lt;/sup&gt;.
        )

    [transformer model] =&gt; Array
        (
            [slug] =&gt; transformer model
            [parent] =&gt; 
            [name] =&gt; transformer model
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; A transformer model operates on {{time series}} or {{sequential data}} using a form of {{attention}}, identifiying past states/tokens that are particualrly related to the current token and then using these in particular as part of predicting future tokens.   This is like the human ability to hear a sentance such as &quot;The cat sat in the deep orange glow of sunset and licked its fur&quot; -- when you read the word 'licked', your mind instantly pulls out the word 'cat' as related and uses that to make sense of the current pont in the sentance.  We use rich {{semantic}} structures to perform this, but transformer models use a {{vector simiilarity}} between a 'key' and 'query' for each token, where the key models the kind of thing that the input token is,  and the query the kind of thing it would like to connect with.
        )

    [transformer models] =&gt; Array
        (
            [slug] =&gt; transformer models
            [parent] =&gt; 
            [name] =&gt; transformer models
            [canonical] =&gt; transformer models
            [see] =&gt; 
            [description] =&gt; 
        )

    [transformer networks] =&gt; Array
        (
            [slug] =&gt; transformer networks
            [parent] =&gt; 
            [name] =&gt; transformer networks
            [canonical] =&gt; transformer models
            [see] =&gt; 
            [description] =&gt; 
        )

    [transition network] =&gt; Array
        (
            [slug] =&gt; transition network
            [parent] =&gt; 
            [name] =&gt; transition network
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; 
            [subentries] =&gt; Array
                (
                    [transition network!augmented] =&gt; Array
                        (
                            [slug] =&gt; transition network!augmented
                            [parent] =&gt; transition network
                            [name] =&gt; augmented
                            [canonical] =&gt; 
                            [see] =&gt; augmented transition network
                            [description] =&gt; 
                        )

                )

        )

    [transition networks] =&gt; Array
        (
            [slug] =&gt; transition networks
            [parent] =&gt; 
            [name] =&gt; transition networks
            [canonical] =&gt; transition network
            [see] =&gt; 
            [description] =&gt; 
        )

    [transition network!augmented] =&gt; Array
        (
            [slug] =&gt; transition network!augmented
            [parent] =&gt; transition network
            [name] =&gt; augmented
            [canonical] =&gt; 
            [see] =&gt; augmented transition network
            [description] =&gt; 
        )

    [transparency] =&gt; Array
        (
            [slug] =&gt; transparency
            [parent] =&gt; 
            [name] =&gt; transparency
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; 
        )

    [transparent] =&gt; Array
        (
            [slug] =&gt; transparent
            [parent] =&gt; 
            [name] =&gt; transparent
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; 
        )

    [transpose] =&gt; Array
        (
            [slug] =&gt; transpose
            [parent] =&gt; 
            [name] =&gt; transpose
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; The transpose of a {{matrix}} M (often written M&lt;sup&gt;T&lt;/sup&gt;) is the matrix reflected along its diagonal. That is given a matrix M&lt;sub&gt;jk&lt;/sub&gt;, its transpose is M&lt;sup&gt;T&lt;/sup&gt;&lt;sub&gt;st&lt;/sub&gt; = M&lt;sub&gt;ts&lt;/sub&gt;.   For example. &lt;br&gt;&lt;br&gt;&lt;table align=&quot;center&quot;&gt; &lt;tr&gt;&lt;th colspan=3 align=center&gt;matrix M&lt;/th&gt;&lt;th&gt;&amp;nbsp;&amp;nbsp;&amp;nbsp;&lt;/th&gt;&lt;th colspan=3 align=center&gt;transpose M&lt;sup&gt;T&lt;/sup&gt;&lt;/th&gt;&lt;/tr&gt; &lt;tr&gt;&lt;td&gt;1&lt;/td&gt;&lt;td&gt;2&lt;/td&gt;&lt;td&gt;3&lt;/td&gt;&lt;td&gt;&lt;td&gt;&lt;td&gt;1&lt;/td&gt;&lt;td&gt;4&lt;/td&gt;&lt;td&gt;7&lt;/td&gt;&lt;/tr&gt;  &lt;tr&gt;&lt;td&gt;4&lt;/td&gt;&lt;td&gt;5&lt;/td&gt;&lt;td&gt;6&lt;/td&gt;&lt;td&gt;&lt;td&gt;&lt;td&gt;2&lt;/td&gt;&lt;td&gt;5&lt;/td&gt;&lt;td&gt;8&lt;/td&gt;&lt;/tr&gt; &lt;tr&gt;&lt;td&gt;7&lt;/td&gt;&lt;td&gt;8&lt;/td&gt;&lt;td&gt;9&lt;/td&gt;&lt;td&gt;&lt;td&gt;&lt;td&gt;3&lt;/td&gt;&lt;td&gt;6&lt;/td&gt;&lt;td&gt;9&lt;/td&gt;&lt;/tr&gt; &lt;/table&gt;
        )

    [travelling salesman problem] =&gt; Array
        (
            [slug] =&gt; travelling salesman problem
            [parent] =&gt; 
            [name] =&gt; travelling salesman problem
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; The travelling salesman problem is a common problem with both practical applicatiosn and used as a challenge for optimisation techniques.  A salesperson needs to visit certain locations that are linked by roads or railways, with different travelling costs or times associated wut them.  The task is to plan a route that visits each taregt location, but with the minimum cost for the travel.  The full solution is known to be {{NP complete}}, that is effectively impossible, but various {{statistical techniques}} and {{machine learning}} may achieve near optimal performance in most real cases.
        )

    [trend adjustments] =&gt; Array
        (
            [slug] =&gt; trend adjustments
            [parent] =&gt; 
            [name] =&gt; trend adjustments
            [canonical] =&gt; 
            [see] =&gt; trend removal
            [description] =&gt; 
        )

    [trend removal] =&gt; Array
        (
            [slug] =&gt; trend removal
            [parent] =&gt; 
            [name] =&gt; trend removal
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; Trend removal (or trend adjustments) try to separate long terms trends from shorter term fluctuations in data.  Typically the long-term trend is estimated, often as a linear growth or decay and then subtracted from the data values.  The term is usually applied to {{time series data}}, but can be appled equally to other forms of {{sequential data}}.  Some transformations implicitly remove trend.  For example taking successive differences of a time series (t&lt;sub&gt;i+1&lt;/sub&gt;&amp;minus;t&lt;sub&gt;i&lt;/sub&gt;) effectively removes any linear trend.  Simlarly, taking successive ratios  (t&lt;sub&gt;i+1&lt;/sub&gt;/t&lt;sub&gt;i&lt;/sub&gt;) removes {{exponential}} trend.  
        )

    [trigger] =&gt; Array
        (
            [slug] =&gt; trigger
            [parent] =&gt; 
            [name] =&gt; trigger
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; In an {{agent}} or other software system, an {[event}} can trigger behaviour such as a {{method}} being executed or a {{message}} being sent.  The term trigger is sometimes used as a noun to refer to the condition that gives rise to the action (e.g. &quot;the trigger for turning on the hearing is when the temperature is below 5 degrees&quot;&quot;)  and sometimes as a verb for the invoking of the action (e.g. &quot;the temperature drop triggered the hearting system to start&quot;).
        )

    [triggers] =&gt; Array
        (
            [slug] =&gt; triggers
            [parent] =&gt; 
            [name] =&gt; triggers
            [canonical] =&gt; trigger
            [see] =&gt; 
            [description] =&gt; 
        )

    [triplestores] =&gt; Array
        (
            [slug] =&gt; triplestores
            [parent] =&gt; 
            [name] =&gt; triplestores
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; Triplestores are a form of {{database}} used to store {{semantic web}} data.. {{RDF}} is encoded as {{subject-predicate-object triples}} and the triplestore stores and indexes these triples in order to make them easy to query and process.  Often {{SPARQL}} is used to query a triplestore.  The {{quadstore}} is an extension of the triplestore to store multiple graphs, that is collections of triples with a common origin or use.
        )

    [trolley problem] =&gt; Array
        (
            [slug] =&gt; trolley problem
            [parent] =&gt; 
            [name] =&gt; trolley problem
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; The trolley problem is a thought experiment in {{moral philosophy}}, but has begun to have practical consequences as {{autonomous vehicles}} become common.  Note that the word 'trolley' in the title is used to descibe a small train in some countries.  The trrain is imagiend to be out of control and about to hit a group of (say five) people.  However, tere is a set of pounts that you can control which will send the train down a sidetrack avoiding the people. Unfortunately there is another person on that track.  You have to decide whether to sacrifice this person to save the greater number on the mainline.
        )

    [truth maintenance system] =&gt; Array
        (
            [slug] =&gt; truth maintenance system
            [parent] =&gt; 
            [name] =&gt; truth maintenance system
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; A truth maintenance system (TMS) is a form of {{non-monotonic reasoning}}.  A TMS links potential assertions (such as 'it is cold' and 'it is winter') and attempts to maintain a coherent set of beliefs about theor truth.  Some are given as facts, others may have their truth determined by assuming defaults where facts are unknown (for example if it is winter one assumes it is also cold).  As new facts arise some of the previous truth values may need to change, for example a temperature reading may show it is avctually an unusually warm wonter's day and hence the truth assigned to 'it is cold' changes. 
        )

    [truth values] =&gt; Array
        (
            [slug] =&gt; truth values
            [parent] =&gt; 
            [name] =&gt; truth values
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; 
        )

    [turing test] =&gt; Array
        (
            [slug] =&gt; turing test
            [parent] =&gt; 
            [name] =&gt; Turing test
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; The Turing test, originally called the Imitation game, was a hypothetical test envisaged by {{Alan Turing}} in 1950 well before computers had the capability of even rudimentary artificial intelligent.  Turing imagined talking to either a computer or a human being at the other end of a text connection.  If it is not possible to tell which is which, then the computer passes the test and can be deemed to be intelligent.  There is debate about whether this really does constitute intelligence, especially now as {{chatbots}} based on {{large language modles}} such as {{chatGPT}} seem to be able to pass the test.
        )

    [turing, alan] =&gt; Array
        (
            [slug] =&gt; turing, alan
            [parent] =&gt; 
            [name] =&gt; Turing, Alan
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; Alan Turing was a mathematician.  He is famous in part for his work on cryptography during the Second World War including designing the {{Bombe}} to help decrypt {{Enigma machine}} messages.  He was also a theoretician about the principles of computation leading to both the concept of the {{Turing machine}}, a form of universal computation, and positing of the 'Imitation game', otherwise known as the {{Turing test}} to ascertain whether a machine is truly intelligent.
        )

    [uncertain reasoning] =&gt; Array
        (
            [slug] =&gt; uncertain reasoning
            [parent] =&gt; 
            [name] =&gt; uncertain reasoning
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; 
        )

    [uncertainty] =&gt; Array
        (
            [slug] =&gt; uncertainty
            [parent] =&gt; 
            [name] =&gt; uncertainty
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; 
        )

    [unconscious bias] =&gt; Array
        (
            [slug] =&gt; unconscious bias
            [parent] =&gt; 
            [name] =&gt; unconscious bias
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; Unconscious bias rerefrs to the way that as humans we make all sorts of assumptions about people based on class, ethnicity, gender or other factors such as how they speak or where they studied.  This is opposed to more conscious {{bias}}, such as explciitly not liking people from a certain country or town based on historic enmity.  Unconscious bias may be based on cultural attitudes that we have implicitly learnt (from media, family or friends) or personal experiences (for example, having had a bad experience with someone of a partcular religion).  The boundary between learning and bias is thus often unclear, but we make choices as individuals and socities as to which are acceptable (e.g. refusing to serve someone acting agressivley) or not (e.g. doing the smae based on the coluor of their skin).  When a {{machine learnng}} system is trisned using pfrevious human decisons or behaviours, it may learn from human {{bias}}, however this is not the only source of  {{bias}} in computer systems.
        )

    [underdetermined] =&gt; Array
        (
            [slug] =&gt; underdetermined
            [parent] =&gt; 
            [name] =&gt; underdetermined
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; The weights or parameters in a {{machine learning}} system are underdetermined if there are many different configurations of parameters that all explain the {{training data}}.  For example, if the system is trying to learn simple arithmatic operations, given an input: (2,2) with expected output (4); both 2+2=4, and 2x2=4; so the operation is underdetermined.  While this is obvious in this small case, the {{hidden layers}} in a {{neural network}} often have millions of weights.  Furthermore, of the layers are {{fully connected}}, then any reordering of the internal nodes and weights would geberate equivalent outputs.  This can also occur in {{statistical methods}} such as {{multilinear regression}} is there are too many {{features}}, any of which can contribute.&lt;br&gt;&lt;br&gt;Underdetermined models may {{overlearn}} from the training data leading to poor {{generalisation}} or {{bias}}.  Typically also they can lead to longer training times as the learning algorithm needs to 'decide' between many competing, but equally good, explanations of the data, and so effectively shifts back and forth between them. 
        )

    [uniform distribution] =&gt; Array
        (
            [slug] =&gt; uniform distribution
            [parent] =&gt; 
            [name] =&gt; uniform distribution
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; A uniform distribution is a {{probability distribution}} where the probabilty of each outcome is equal.
        )

    [uniform resource identifiers] =&gt; Array
        (
            [slug] =&gt; uniform resource identifiers
            [parent] =&gt; 
            [name] =&gt; uniform resource identifiers
            [canonical] =&gt; 
            [see] =&gt; uri
            [description] =&gt; 
        )

    [uniform resource locators] =&gt; Array
        (
            [slug] =&gt; uniform resource locators
            [parent] =&gt; 
            [name] =&gt; uniform resource locators
            [canonical] =&gt; 
            [see] =&gt; url
            [description] =&gt; 
        )

    [uniform sampling rate] =&gt; Array
        (
            [slug] =&gt; uniform sampling rate
            [parent] =&gt; 
            [name] =&gt; uniform sampling rate
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; A uniform sampling rate is when a {{time series}} is sampled at fixed intervals, say once a second.  The majority of {{time series anaysis}} techniques assume a uniform sampling rate. However, there are many knds of data where this is not the case, for example, medical data, such as blood results, only occur when a patient visits a doctor or while in hospital the doctir decides a further test is needed; thus they are thus both {{sporadic}} and almost cedrtainly at times that are not typical of times when no sample is taken.  Even where the intention is to have a uniform sampling rate, sensor failures or lost data may mean there are {{missing values}}.
        )

    [unintended bias] =&gt; Array
        (
            [slug] =&gt; unintended bias
            [parent] =&gt; 
            [name] =&gt; unintended bias
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; 
        )

    [unique identifier] =&gt; Array
        (
            [slug] =&gt; unique identifier
            [parent] =&gt; 
            [name] =&gt; unique identifier
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; In a {{database}} or other {{infoirnation system}}, a unique identifier for a data item or record is a combination of features of the data or a generated identifier that is not shared by any other item in the dataset.   For example, in a geographic information system, we might use the physical location of a town as a identifier rather than name, as there are often many towns with the same name.  Often these feature-based identifications are difficult to collect or maintain, or don't even exist, hence the frequent use of generated identifiers such as health or tax numers for people.  Idetifiers may be unique in a single database (e.g. record number) or may be a {{global unique idientifer}} ({GUID}} such as {{URI}}s for the web.
        )

    [unsupervised] =&gt; Array
        (
            [slug] =&gt; unsupervised
            [parent] =&gt; 
            [name] =&gt; unsupervised
            [canonical] =&gt; unsupervised learning
            [see] =&gt; 
            [description] =&gt; 
        )

    [unsupervised learning] =&gt; Array
        (
            [slug] =&gt; unsupervised learning
            [parent] =&gt; 
            [name] =&gt; unsupervised learning
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; In unsupervised learning the {{training set}} is {{unlabelled}}, that is it has no markers to indicate a desired grouping or value.  Unsupervised learning algorithms look for commonalities to group or infer other properties of the data.  This may be used on its own, or be used as the first stage before human labelling or for a subsequent {{supervised learning}} method.
        )

    [upper bound] =&gt; Array
        (
            [slug] =&gt; upper bound
            [parent] =&gt; 
            [name] =&gt; upper bound
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; An upper bound is an estimate, {{heuristic}} or calculated value that you know lies at or above the largest possible value of an unknown value or function. For example, before you roll a pair of dice, you know that the &lt;em&gt;upper bound&lt;/em&gt; of the sum of the dice scores will be 36 (both dice show six dots).  Knowing an upper bound can sometimes allow whole branches of a {{search tree}} or similar structure to be {{pruned}}.  For example, if you are looking for the largest value in a tree and have found a node with value 7, and then consider an alterantive branch in the tree where a {{heuristic}} says the {{upper bound}} of all items below is 5, you can safely ignore the whole sub-tree. &lt;br&gt;&lt;br&gt;. the opposite of an upper bound is a {{lower bound}}.
        )

    [uri] =&gt; Array
        (
            [slug] =&gt; uri
            [parent] =&gt; 
            [name] =&gt; URI
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; A URI (uniform resource identifier) is an extension of the URL.  Every URL is a URI, but not every URI can be accessed in the way a URL can.  It has the same format as URL, but can also refer to non-information entities such as people or places.  The URI acts as a way of refering to the entity so that different information repositories or databases know they are refering to the same thing.  For example https://alandix.com/uri/aibook might be used as a way of talking about the paper book.  This is used in particular in {{RDF}} and other {{semantic web}} techologies.
        )

    [url] =&gt; Array
        (
            [slug] =&gt; url
            [parent] =&gt; 
            [name] =&gt; URL
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; A URL (uniform resource locator) is the familar address in the top bar of a {{web browser}}.  For example, https://alandix.com/glossary/aibook is the URL of this glossary.Unlike earlier {{hypertext}} a URL is both globally unique  and can be accessed to give the resource (most commonly a web page or image).
        )

    [user event] =&gt; Array
        (
            [slug] =&gt; user event
            [parent] =&gt; 
            [name] =&gt; user event
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; A user event in a {{user interface}} or when a human communicates with an {{agent}}, is when the user initiates some form of action or message, for example asking a question to an {{agent}} or clicking a mouse in a {{user interface}}.  This is in contrast to a {{system event}}.
        )

    [user events] =&gt; Array
        (
            [slug] =&gt; user events
            [parent] =&gt; 
            [name] =&gt; user events
            [canonical] =&gt; user event
            [see] =&gt; 
            [description] =&gt; 
        )

    [user interface] =&gt; Array
        (
            [slug] =&gt; user interface
            [parent] =&gt; 
            [name] =&gt; user interface
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; 
        )

    [validation rules] =&gt; Array
        (
            [slug] =&gt; validation rules
            [parent] =&gt; 
            [name] =&gt; validation rules
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; 
        )

    [value network] =&gt; Array
        (
            [slug] =&gt; value network
            [parent] =&gt; 
            [name] =&gt; value network
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; 
        )

    [variable-order markov models] =&gt; Array
        (
            [slug] =&gt; variable-order markov models
            [parent] =&gt; 
            [name] =&gt; variable-order markov models
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; While a simple {{Markov model}}  only uses the last state/token to predict the next one, a variable-order Markov model may use several past states.  For example a third-order Markov model, gives the probabilty of the state S&lt;sub&gt;t&lt;/sub&gt; as a function of the last three states: &lt;br&gt; &amp;nbsp; &amp;nbsp; &amp;nbsp; P(S&lt;sub&gt;t&lt;/sub&gt;=s) &amp;nbsp;  = &amp;nbsp;  f(s&lt;sub&gt;t-1&lt;/sub&gt;,s&lt;sub&gt;t-2&lt;/sub&gt;,s&lt;sub&gt;t-3&lt;/sub&gt;)
        )

    [variance] =&gt; Array
        (
            [slug] =&gt; variance
            [parent] =&gt; 
            [name] =&gt; variance
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; 
        )

    [vcard ontology] =&gt; Array
        (
            [slug] =&gt; vcard ontology
            [parent] =&gt; 
            [name] =&gt; vCard ontology
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; The vCard ontology is a standard way of describing contact details for people ... that is the information one might find in a 'visiting card'.  It arose as an {{emergent standard}} for interchange between contact management systems and as a web {{microformat}}.  The {{RDF}} form is used extensively in {{semantic web}} {{ontologies}}.
        )

    [vergence] =&gt; Array
        (
            [slug] =&gt; vergence
            [parent] =&gt; 
            [name] =&gt; vergence
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; In a {{stereo camera}}, the vergence is the horizontal angle between two cameras in the head, which allows both cameras to fixate on one object and also can be used to gauge the object's distance.
        )

    [version spaces] =&gt; Array
        (
            [slug] =&gt; version spaces
            [parent] =&gt; 
            [name] =&gt; version spaces
            [canonical] =&gt; version-space algorithm
            [see] =&gt; 
            [description] =&gt; 
        )

    [version-space] =&gt; Array
        (
            [slug] =&gt; version-space
            [parent] =&gt; 
            [name] =&gt; version-space
            [canonical] =&gt; version-space algorithm
            [see] =&gt; 
            [description] =&gt; 
        )

    [version-space algorithm] =&gt; Array
        (
            [slug] =&gt; version-space algorithm
            [parent] =&gt; 
            [name] =&gt; version-space algorithm
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; 
        )

    [vilfredo pareto] =&gt; Array
        (
            [slug] =&gt; vilfredo pareto
            [parent] =&gt; 
            [name] =&gt; Vilfredo Pareto
            [canonical] =&gt; 
            [see] =&gt; pareto, vilfredo
            [description] =&gt; 
        )

    [visual analytics] =&gt; Array
        (
            [slug] =&gt; visual analytics
            [parent] =&gt; 
            [name] =&gt; visual analytics
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; Visual analytics combines {{visualisation}}, {{machine learning}} and {{statistical analysis}} in an interactuve system to aid human {{sense making}} and {{decision making}} especially when delaing with complex data.
        )

    [visualisation] =&gt; Array
        (
            [slug] =&gt; visualisation
            [parent] =&gt; 
            [name] =&gt; visualisation
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; 
        )

    [visualising data] =&gt; Array
        (
            [slug] =&gt; visualising data
            [parent] =&gt; 
            [name] =&gt; visualising data
            [canonical] =&gt; 
            [see] =&gt; visualisation
            [description] =&gt; 
        )

    [voxel] =&gt; Array
        (
            [slug] =&gt; voxel
            [parent] =&gt; 
            [name] =&gt; voxel
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; A voxel is the 3D equivalent of a {{pixel}}, that is a small rectangular (usually cubic) patch of space.  For example, if a room is 5 metres long, by 4 metres wide, by 3metres high (5mx4mx3m), we might divide it into 5000x4000x3000 1mm cubes for the purposes of indexing objects or performing 3D rendering.  While photographs are usually 2D, various forms of medical scanners, such as ultrasound or MRI generate 3D views of the patient.  However, these may have non-cubic voxels as they create a series of frames or slices and the gap between slices (usually deterimined by mechanical movement of the device or patient) may be different from the resolution within a slice.
        )

    [waltz's algorithm] =&gt; Array
        (
            [slug] =&gt; waltz's algorithm
            [parent] =&gt; 
            [name] =&gt; Waltz's algorithm
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; 
        )

    [warren, d] =&gt; Array
        (
            [slug] =&gt; warren, d
            [parent] =&gt; 
            [name] =&gt; Warren, D
            [canonical] =&gt; warren, david
            [see] =&gt; 
            [description] =&gt; 
        )

    [watson] =&gt; Array
        (
            [slug] =&gt; watson
            [parent] =&gt; 
            [name] =&gt; Watson (IBM)
            [canonical] =&gt; 
            [see] =&gt; ibm watson
            [description] =&gt; 
        )

    [wavelength] =&gt; Array
        (
            [slug] =&gt; wavelength
            [parent] =&gt; 
            [name] =&gt; wavelength
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; The wavelength is the peak-to-peak distance of a wave.  For sea waves this may vary form a few centimetres for ripple to may metres for deep sea waves.  For sounds this varies from a few centimetres for notes at the top of a piano to several metres for those at the bottom.  For light the wavelength is between around 400nm (violet) to 700nm (red).   This is to be contrasted with the {{frequency}}, which os the time taken for the wave to move for. peak-to-peak at a fixed location.  For sound waves the two are relatred by the speed of sound, for electriomagnetic waves the speed of lught: &lt;br&gt; &amp;nbsp; &amp;nbsp; &amp;nbsp; wavelength / frequency &amp;nbsp;  =  &amp;nbsp; speed
        )

    [wavelet] =&gt; Array
        (
            [slug] =&gt; wavelet
            [parent] =&gt; 
            [name] =&gt; wavelet
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; A wavelet is a small function, which has finite/local extent in time or space.  For example: &lt;br&gt; &amp;nbsp; &amp;nbsp; &amp;nbsp;  f&lt;sub&gt;s(t)  &amp;nbsp;= &amp;nbsp; 1 - |t-s|  &amp;nbsp; if (t-s) &lt; 1 ,  &amp;nbsp; =  &amp;nbsp;0 otherwise &lt;b&gt;&lt;br&gt; Wavletes can be one diennsional, for use over {{time series}} series or {{sequential data}}, 2D for images, or higher dimensions.   Wavelets usually form part fo a collection with different wavelests at different sizes and locations, so that any \term{time series} or image can be expressed as a sum of the wavekets in the collection -- called a {{wavelet transform}} of the function.
        )

    [wavelets] =&gt; Array
        (
            [slug] =&gt; wavelets
            [parent] =&gt; 
            [name] =&gt; wavelets
            [canonical] =&gt; wavelet
            [see] =&gt; 
            [description] =&gt; 
        )

    [wavelet transform] =&gt; Array
        (
            [slug] =&gt; wavelet transform
            [parent] =&gt; 
            [name] =&gt; wavelet transform
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; A wavelet transform decomposes a function or time series into a sum of products of {{wavelets}} from a collection.  The collection is often defined mathematically using a {{kernal}}, which then mans thet thay have good properties.  It is rather like a {{Fourier transform}}, as the wavelets typically opertae at different tmeoral or spatial scals, but whereas the sine/cos funcitons used in the {{Fourier transform}} extend over the etire temproal or spatial domain, wavelets are local.
        )

    [wavelet transformation] =&gt; Array
        (
            [slug] =&gt; wavelet transformation
            [parent] =&gt; 
            [name] =&gt; wavelet transformation
            [canonical] =&gt; wavelet transform
            [see] =&gt; 
            [description] =&gt; 
        )

    [wavelet transforms] =&gt; Array
        (
            [slug] =&gt; wavelet transforms
            [parent] =&gt; 
            [name] =&gt; wavelet transforms
            [canonical] =&gt; wavelet transform
            [see] =&gt; 
            [description] =&gt; 
        )

    [weak ai] =&gt; Array
        (
            [slug] =&gt; weak ai
            [parent] =&gt; 
            [name] =&gt; weak AI
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; Proponents of weak AI take the position that it is sufficent if AI creates computer systems that perform in an intelligent manner even if they do so in very diffeernt ways to humans.  That is they are not trying to deeply emulate human intelligence in the way that proponents of {{strong AI}} aim to do.
        )

    [weak learners] =&gt; Array
        (
            [slug] =&gt; weak learners
            [parent] =&gt; 
            [name] =&gt; weak learners
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; In {{machine learniong}} weak learners are sub-models that are in themselves not that good, but can potentially be combined, or modified in someway to make them stringer.  For example, {{random forest}} techniques create large numbers of {[decision trees}} each of which may not be that effective, but when combined by higher-level decision algorithm (such as weighted voting), have proven very powerful.  
        )

    [web of data] =&gt; Array
        (
            [slug] =&gt; web of data
            [parent] =&gt; 
            [name] =&gt; web of data
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; The {{world wide web}} that we browse consists of human rewadable web paes.  However, the web was originallyconceived in {{CERN}} as a way to share data.  {{Tim Berners-Lee}}'s vision is for a &lt;em&gt;web of data&lt;/em&gt; that is machine readable, so that automated systems and reasoning can apply at a global scale. Central to this vision is the {{semantic web}}, which means that not only the data itself, but also the {{meta-information}} necessary to interpret it are available in a machine readable and linkable fashion.
        )

    [web ontology language] =&gt; Array
        (
            [slug] =&gt; web ontology language
            [parent] =&gt; 
            [name] =&gt; web ontology language
            [canonical] =&gt; 
            [see] =&gt; owl
            [description] =&gt; 
        )

    [web science] =&gt; Array
        (
            [slug] =&gt; web science
            [parent] =&gt; 
            [name] =&gt; web science
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; Web Science is a cross-disciplinary field studying the web as a wholistic entity including humans and machines.  The web could simply be seen as a graph of interconnected pages and other data.  Web science recognises that this data of the web is part of a larger system that also includes people interacting with and through the web, and physiclal processes being monitored to generate web data and also being controlled through the web.  Web science thefore interacts with many fields including {{data science}}, {{human computer interaction}} and the {{internet of things}}
        )

    [web scraping] =&gt; Array
        (
            [slug] =&gt; web scraping
            [parent] =&gt; 
            [name] =&gt; web scraping
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; Web scraping involves taking web pages intended for human clnsumption and extracting key data from it, using the structure of the page for some form of {{template matching}} or similar process.  For example, some of the data elemints on the page may have specific CSS classes that can be used to work out where they are and then extract their data.  The output of web scraping is data that can then be used for other processing algorithms.
        )

    [weizenbaum, joseph] =&gt; Array
        (
            [slug] =&gt; weizenbaum, joseph
            [parent] =&gt; 
            [name] =&gt; Weizenbaum, Joseph
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; Joseph Weizenbaum was an early researcher and theorist in AI and in particular created the chatbot {{ELIZA}}.  Weizenbaum was a sceptic about the ultimate potential for computers to be truly intelligent.
        )

    [white matter] =&gt; Array
        (
            [slug] =&gt; white matter
            [parent] =&gt; 
            [name] =&gt; white matter
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; White matter is the material in the centre of the brain as compated with the {{grey matter}} at the outside.  The white matter that is principally composed of {{axons}}, the nerve fibres that connect the bodies of {{neorones}} that are concentretd in the {{grey matter}}.
        )

    [windowing] =&gt; Array
        (
            [slug] =&gt; windowing
            [parent] =&gt; 
            [name] =&gt; windowing
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; Many techniques for {{time-series}} or {{sequential data}} work by taking a fixed size window, that is the last N items of the series, and then use each wondow as if it were separate data, eiethre for training ro execution.  Examples include {{moving average}} methods in {{time series analysis}}, simple {{Markov models}} and applying {{neural networks}} to the data windows.  Wondowing methods are always {{finite impluse response}} as data outside of the wondow cannot affect the current response.
        )

    [winner takes all] =&gt; Array
        (
            [slug] =&gt; winner takes all
            [parent] =&gt; 
            [name] =&gt; winner takes all
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; Many classification or decsiion making algorithms output a set of scores, or weights for each potential outcome rather than a single classification or decision.  A winner takes all strategy takes the highest scrore and uses that to make a single choice.  This is a low risk stregtegy, but often means that systems do not sufficiently explore potential behaviours.   Alternative strategies include using the scores to create a probability and then randomly make the final decision based on that.
        )

    [winograd, terry] =&gt; Array
        (
            [slug] =&gt; winograd, terry
            [parent] =&gt; 
            [name] =&gt; Winograd, Terry
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; Terry Winograd is one of the major figures in the development of AI and in particular created  the blocks world {{SHRLDU}}.
        )

    [wisdom of the crowds] =&gt; Array
        (
            [slug] =&gt; wisdom of the crowds
            [parent] =&gt; 
            [name] =&gt; wisdom of the crowds
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; The wisdom of the crowds is the thesis that if you ask sufficient people a question then even if each is individually not ver knowledgable, the consensus view is remarkably accurate.  This seems to work for certain kinds of problem where tno expert knowledge is required, but if, say, the question concerned qunatum string theory, it seems less lilely that a random crowd would be helpful unless it happend to be a crowd of quantum string theorists.  However, a computatuonal variant of this princple does seem to work well in {{ensemble}} methods in AI.
        )

    [wizard] =&gt; Array
        (
            [slug] =&gt; wizard
            [parent] =&gt; 
            [name] =&gt; wizard
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; A wizard is a part of the user interface in a system that helps a user confugure or otherwsie adjust the system usually using simple question-answer or form-filling interactions.  Wizards tend to be pre-programmed for very specific aspects of a system, but there is the potential for intelligent assustants to be far more flexible.
        )

    [wizards] =&gt; Array
        (
            [slug] =&gt; wizards
            [parent] =&gt; 
            [name] =&gt; wizards
            [canonical] =&gt; wizard
            [see] =&gt; 
            [description] =&gt; 
        )

    [word vector] =&gt; Array
        (
            [slug] =&gt; word vector
            [parent] =&gt; 
            [name] =&gt; word vector
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; A word vector is the value associated with a word in {{word2vec}}.  It captures some level of meaning, for example, if WV('dog'), WV('puppy'), WV('cat') and WV('kitten') are the word vectors assocated with 'dog', 'puppy', 'cat' and 'kitten' respectvely, then the vector difference WV('dog')&ndash;WV('puppy') will typically be close to WV('cat')&ndash;WV('kitten').  Tis makes it possible to asnwre questions such as &quot;dog is to puppy as cat is to what?&quot;
        )

    [word2vec] =&gt; Array
        (
            [slug] =&gt; word2vec
            [parent] =&gt; 
            [name] =&gt; word2vec
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; word2vec creates a vector representation of a word that captures certaib aspects of the menaing of the word.  It works by first creating a 'Skip-gram model' that predicts the likelihood of curroundng words, and then extracts an internal layer of this model to form a {{latent space}}.  The value of a word in this latent space is called a {{word vector}}.  The geometry of the {{latent space}} often captures crucially relationships, so that, for example, the vector difference between 'dog' and  'puppy' is similar to that between 'cat' and 'kitten'.  These word vectors can be used in further steps of an {{NLP}} process.
        )

    [world wide web] =&gt; Array
        (
            [slug] =&gt; world wide web
            [parent] =&gt; 
            [name] =&gt; World Wide Web
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; Although we mostly think of the web in terms of human readable web pages, the invention of the {{World Wide Web}} by {{Tim Berners-Lee}} was originally focused on helping physicists at {{CERN}} to share data.  This vision is expressed in the {{semantic web}} a number of technologies aimed at making the knowledge on the web also understandable by machines.  In addition, AI is being used extensively, for example in {{recommender systems}} that suggest items for you in media and shopping web sites, and to celect items for feeds in social media applications. 
 Furthermore, even human readable pages are being used as data for {{machine learning}} notably {{large language models}}.  Indeed, it is this huge volume of {{big data}} available on the web which has been one of the main drivers for the latest wave of AI success.
        )

    [xcon] =&gt; Array
        (
            [slug] =&gt; xcon
            [parent] =&gt; 
            [name] =&gt; XCON
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; XCON was an early {{expert system}} developed by DigitalElectronics Corporation to configure VAX computer systems
        )

    [xor problem] =&gt; Array
        (
            [slug] =&gt; xor problem
            [parent] =&gt; 
            [name] =&gt; XOR problem
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; Early {{single-layer perceptrons}} were not able to solve the XOR problem, that is how to calculate the exclusive or of two inputs.
        )

    [zero-sum game] =&gt; Array
        (
            [slug] =&gt; zero-sum game
            [parent] =&gt; 
            [name] =&gt; zero-sum game
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; This is a game where one player's gain means the other player's loss.   If it is a simple win-lose game, then only one player can win, and eiether the other loses, or they both draw.  In a game with scores, it emasn that the total score is fixed.  {{Noughts and crosses}} is an example of a simple win-lose zero-sum game.  Some {{game playing algorithms}} such as {{alpha-beta pruning}} only work for zero-sum games.    This is the opposite of a {{non-zero-sum game}}.
        )

    [zombie alice] =&gt; Array
        (
            [slug] =&gt; zombie alice
            [parent] =&gt; 
            [name] =&gt; Zombie Alice
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; Zombie Alice is used in thought experiments by philosophers discussing {{consciousness}}.  They imagine zombie Alice wo is just like a human Alice, in all ways except has no real mind.  If Zombie Alice behaves in all respects like a conscious person  are they indeed conscious, or does this show that consciousness is more than behaviour?
        )

    [zoom] =&gt; Array
        (
            [slug] =&gt; zoom
            [parent] =&gt; 
            [name] =&gt; zoom
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; In a camera, the zoom increases magnification and thus allows distant obecst to be viewed more easily.
        )

)


  -->

<!DOCTYPE html PUBLIC "-//W3C//DTD XHTML 1.0 Transitional//EN" "http://www.w3.org/TR/xhtml1/DTD/xhtml1-transitional.dtd">
<html xmlns="http://www.w3.org/1999/xhtml">
<head>
<meta http-equiv="Content-Type" content="text/html; charset=UTF-8" />
<title>Artificial Intelligence</title>
<meta name="twitter:card" content="summary" />
<meta name="twitter:site" content="@alanjohndix" />
<meta name="twitter:creator" content="@alanjohndix" />
<meta property="og:url" content="https://alandix.com/glossary/aibook" />
<meta property="og:title" content="Artificial Intelligence" />
<meta property="og:description" content="Terms from Artificial Intelligence: humans at the heart of algorithms" />
<meta property="og:image" content="https://alandix.com/glossary/assets/images/aibook-icon-128x128.png" />
<link rel="icon" href="https://alandix.com/glossary/assets/images/aibook-icon-128x128.png" />
<script
  src="https://code.jquery.com/jquery-3.4.1.min.js"
  integrity="sha256-CSXorXvZcTkaix6Yvo6HppcZGetbYMGWSFlBw8HfCJo="
  crossorigin="anonymous"></script>
<link rel="stylesheet" href="https://code.jquery.com/ui/1.12.0/themes/smoothness/jquery-ui.css">
<link rel="stylesheet" href="https://alandix.com/glossary/assets/css/style.css">
<script src="https://code.jquery.com/ui/1.12.1/jquery-ui.js"></script>
<script src="https://alandix.com/glossary/assets/js/main.js"></script>
<script type="text/javascript" src="https://alandix.com/code/favicon/favicon.js"></script>
<script language="javascript">
$(function() {
  loadFavicons(document);
});
</script></style>
<script>
var path_info = "aibook";
var glossaries = {"hcistats":{"slug":"hcistats","name":"HCI Statistics","description":"Terms from <a href=\"https:\/\/alandix.com\/statistics\/book\/\">Statistics for HCI: Making Sense of Quantitative Data<\/a>","list-message":"Terms in <strong>bold<\/strong> have additional details (short description, links, etc.).  <br>Those in <em>italics<\/em> are stubs at present, just pointing to relevant pages in the book.","item-message":"","icon":"cover-stats-for-HCI-square-128x128.jpg","twitter_site":"@HciStatistics","url":"https:\/\/alandix.com\/statistics\/book\/","menu":[{"name":"book","url":"https:\/\/alandix.com\/statistics\/book\/"}],"options":{"data_api":"https:\/\/datatodata.org\/HCIstats\/data.php\/"}},"aibook":{"slug":"aibook","name":"Artificial Intelligence","description":"Terms from <a href=\"https:\/\/alandix.com\/aibook\/\">Artificial Intelligence: humans at the heart of algorithms<\/a>","list-message":"Terms in <strong>bold<\/strong> have additional details (short description, links, etc.).  <br>Those in <em>italics<\/em> are stubs at present, just pointing to relevant pages in the book.","item-message":"","icon":"aibook-icon-128x128.png","url":"https:\/\/alandix.com\/aibook\/","options":{"json_entries":"\/var\/www\/vhosts\/alandix.com\/httpdocs\/glossary\/private\/data\/aibook-info.json"},"menu":[{"name":"book","url":"https:\/\/alandix.com\/aibook\/"}]}};
var glossary_info = {"slug":"aibook","name":"Artificial Intelligence","description":"Terms from <a href=\"https:\/\/alandix.com\/aibook\/\">Artificial Intelligence: humans at the heart of algorithms<\/a>","list-message":"Terms in <strong>bold<\/strong> have additional details (short description, links, etc.).  <br>Those in <em>italics<\/em> are stubs at present, just pointing to relevant pages in the book.","item-message":"","icon":"aibook-icon-128x128.png","url":"https:\/\/alandix.com\/aibook\/","options":{"json_entries":"\/var\/www\/vhosts\/alandix.com\/httpdocs\/glossary\/private\/data\/aibook-info.json"},"menu":[{"name":"book","url":"https:\/\/alandix.com\/aibook\/"}]};
var glossary = "aibook";
var term_info = null;
var term = null;
</script>
<script>window.twttr = (function(d, s, id) {
  var js, fjs = d.getElementsByTagName(s)[0],
    t = window.twttr || {};
  if (d.getElementById(id)) return t;
  js = d.createElement(s);
  js.id = id;
  js.src = "https://platform.twitter.com/widgets.js";
  fjs.parentNode.insertBefore(js, fjs);

  t._e = [];
  t.ready = function(f) {
    t._e.push(f);
  };

  return t;
}(document, "script", "twitter-wjs"));</script></head>

<body class="glossary glossary-aibook">
<div id="fb-root"></div>
<script async defer crossorigin="anonymous" src="https://connect.facebook.net/en_GB/sdk.js#xfbml=1&version=v6.0&appId=212157625472599"></script>
<header>
<img src="https://alandix.com/glossary/assets/images/aibook-icon-128x128.png" class="icon"><h1>Artificial Intelligence</h1>
<h2><span class="description">Terms from <a href="https://alandix.com/aibook/">Artificial Intelligence: humans at the heart of algorithms</a></span>
</h2></header>
<nav>
<ul class="topmenu">
<li><span class="inactive">glossary</span></li><li><a href="https://alandix.com/aibook/" class="active">book</a></li></ul>
<ul class="social">
<li class="twitter"><a class="twitter-share-button" href="https://twitter.com/intent/tweet?text=Glossary item">Tweet</a></li>
<li class="facebook"><div class="fb-share-button" data-href="https://alandix.com/glossary/aibook" data-layout="button" data-size="small"><a target="_blank" href="https://www.facebook.com/sharer/sharer.php?u=https%3A%2F%2Fexample.com%2Ftest&amp;src=sdkpreparse" class="fb-xfbml-parse-ignore">Share</a></div></li>
</ul>
</nav>
<div class="content">
<p class="message">Terms in <strong>bold</strong> have additional details (short description, links, etc.).  <br>Those in <em>italics</em> are stubs at present, just pointing to relevant pages in the book.</p>
<ul class="glossary_list">
  <li ><span class="name"><a href="https://alandix.com/glossary/aibook/a* algorithm">A* algorithm</a></span><!--  Array
(
    [slug] =&gt; a* algorithm
    [parent] =&gt; 
    [name] =&gt; A* algorithm
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; 
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/aibook/abductive reasoning">abductive reasoning</a></span><!--  Array
(
    [slug] =&gt; abductive reasoning
    [parent] =&gt; 
    [name] =&gt; abductive reasoning
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; Abductve reasoning considers various explainations for observed situation and tries to find the simplest or most likely.  It is a {{knowledge-rich}} method of reasoning that can make ({{defeasible}}) deductions beyond the incontovertable facts (used by {{deductive reasong}}) and available empirical data (used by {{inducdtuve reasoning}}).  As sush it is a form of {{nin-monotonic reasoning}} as new knowledge could change the potetial explanations.  
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/aibook/accountability">accountability</a></span><!--  Array
(
    [slug] =&gt; accountability
    [parent] =&gt; 
    [name] =&gt; accountability
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; This refers to legal or other means by which companies, governments or individuals are held to account for AI algorithms or systems that they produce or deploy.  If something goes wrong they may be taken to court, fined, imprisoned or at least be subject to public approbation.  It is believed if that the costs of negligence are high enough, then companies and other actors will act responsibly.
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/aibook/accuracy">accuracy</a></span><!--  Array
(
    [slug] =&gt; accuracy
    [parent] =&gt; 
    [name] =&gt; accuracy
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; Accuracy is the difference between the output or prediction of an algorithm and the true value.  For numerical results it is often measured using the {{mean square error}} (the average of the square of the difference) or absolute difference.  For classifications there are several different kinds of measures that are important, including {{precision}} and {{recall}} for binary decisions, so the word 'accuracy can be ambiguous.
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/aibook/actstar">ACT<sup>*</sup></a></span><!--  Array
(
    [slug] =&gt; actstar
    [parent] =&gt; 
    [name] =&gt; ACT&lt;sup&gt;*&lt;/sup&gt;
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; ACT&lt;sup&gt;*&lt;/sup&gt; is a cognitive architecture developed by {{John Robert Anderson}} and others in the mid-1970s.  It and its successor {{ACT-R}} have been highly influential within cognitive science..
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/aibook/act-r">ACT-R</a></span><!--  Array
(
    [slug] =&gt; act-r
    [parent] =&gt; 
    [name] =&gt; ACT-R
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; ACT-R (formally known as {{ACT*}}) has been developed by {{John Robert Anderson}} and others over the past 40 years. It comprises a large long- term memory in the form of a {{semantic net}}, a small working memory of active items and a production system that operates on the memories.   ACT-R includes mechanisms (i) to make commonly used instances of general rules more specific (called {{proceduralisation}}) and also to take multiple specifis rules with similar structure and create a single more abstract rule (called {{generalisation}}).
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/aibook/active vision">active vision</a></span><!--  Array
(
    [slug] =&gt; active vision
    [parent] =&gt; 
    [name] =&gt; active vision
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; Active vision is when the camera or camera platform changes its orientation or position in order to better recogonise or interpret and object or scene.  For example, you might look at an object from several angles, or even peek aorund the corner of a building.  In some cases a similar effect can be achieved by manipulating the object itself if it is held in a robotoc hand, or is on a moving platform.
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/aibook/activity recognition">activity recognition</a></span><!--  Array
(
    [slug] =&gt; activity recognition
    [parent] =&gt; 
    [name] =&gt; activity recognition
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; This refers to situations where an AI system needs to recognise some form of human activity, for example, cooking food in kitchen, or leavinga. Suspicious package at an airport.  Activity recognition systems often have a camera and/or some other form of sensors and have to both identifiy aspects of a single frame or reading, but also the changing pattern over time.
)
  --></li>
  <li ><span class="name">Ada Lovelace</span> - See <span class="has_content"><a href="https://alandix.com/glossary/aibook/lovelace, ada">Lovelace, Ada</a></span><!--  Array
(
    [slug] =&gt; ada lovelace
    [parent] =&gt; 
    [name] =&gt; Ada Lovelace
    [canonical] =&gt; 
    [see] =&gt; lovelace, ada
    [description] =&gt; 
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/aibook/adjacency statistics">adjacency statistics</a></span><!--  Array
(
    [slug] =&gt; adjacency statistics
    [parent] =&gt; 
    [name] =&gt; adjacency statistics
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; In a text or in general a sequence, adjacency statistics measure the frequency with which words or tokens occur next to each other.  For example, in the sequence BAAACBBAAC, the pair (or 2-gram) AA occurs three times, AC and  BA each occur twice, and BB and  CB each occur once.  No three tokens (3-gram) in the example occur more than once, but in longer texts 3-grams (such as the words 'artificial neural network') and longer sequences may recur.
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/aibook/adversarial learning">adversarial learning</a></span><!--  Array
(
    [slug] =&gt; adversarial learning
    [parent] =&gt; 
    [name] =&gt; adversarial learning
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; Adversarial learning/training pitches two or more algorithms against each other in a competitive situation. Each adversary uses {{machine learning}} to attempt to gain advantage over the other.  It takes inspiration from {{zero-sum games}}, where one player's gaims are the other player's loss.   The technique haa been used in {{generative adversarial networks}} for image generation, for example in {{deep fakes}}, where one adversary tries to generate new images and the other to distinguish generated images from real ones.  {{Self play}} is a similar technique in {{game playing}} AI.
)
  --></li>
  <li ><span class="name"><a href="https://alandix.com/glossary/aibook/agents">agents</a></span>- has subentries<ul class="subentries"><li ><span class="name"><a href="https://alandix.com/glossary/aibook/agents!email filtering">email filtering</a></span><!--  Array
(
    [slug] =&gt; agents!email filtering
    [parent] =&gt; agents
    [name] =&gt; email filtering
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; 
)
  --></li>
<li ><span class="name"><a href="https://alandix.com/glossary/aibook/agents!embodiment">embodiment</a></span><!--  Array
(
    [slug] =&gt; agents!embodiment
    [parent] =&gt; agents
    [name] =&gt; embodiment
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; 
)
  --></li>
<li ><span class="name"><a href="https://alandix.com/glossary/aibook/agents!intelligent filtering">intelligent filtering</a></span><!--  Array
(
    [slug] =&gt; agents!intelligent filtering
    [parent] =&gt; agents
    [name] =&gt; intelligent filtering
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; 
)
  --></li>
<li ><span class="name"><a href="https://alandix.com/glossary/aibook/agents!machine learning">machine learning</a></span><!--  Array
(
    [slug] =&gt; agents!machine learning
    [parent] =&gt; agents
    [name] =&gt; machine learning
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; 
)
  --></li>
<li ><span class="name"><a href="https://alandix.com/glossary/aibook/agents!messages">messages</a></span><!--  Array
(
    [slug] =&gt; agents!messages
    [parent] =&gt; agents
    [name] =&gt; messages
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; 
)
  --></li>
<li ><span class="name"><a href="https://alandix.com/glossary/aibook/agents!methods">methods</a></span><!--  Array
(
    [slug] =&gt; agents!methods
    [parent] =&gt; agents
    [name] =&gt; methods
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; 
)
  --></li></ul><!--  Array
(
    [slug] =&gt; agents
    [parent] =&gt; 
    [name] =&gt; agents
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; 
    [subentries] =&gt; Array
        (
            [agents!email filtering] =&gt; Array
                (
                    [slug] =&gt; agents!email filtering
                    [parent] =&gt; agents
                    [name] =&gt; email filtering
                    [canonical] =&gt; 
                    [see] =&gt; 
                    [description] =&gt; 
                )

            [agents!embodiment] =&gt; Array
                (
                    [slug] =&gt; agents!embodiment
                    [parent] =&gt; agents
                    [name] =&gt; embodiment
                    [canonical] =&gt; 
                    [see] =&gt; 
                    [description] =&gt; 
                )

            [agents!intelligent filtering] =&gt; Array
                (
                    [slug] =&gt; agents!intelligent filtering
                    [parent] =&gt; agents
                    [name] =&gt; intelligent filtering
                    [canonical] =&gt; 
                    [see] =&gt; 
                    [description] =&gt; 
                )

            [agents!machine learning] =&gt; Array
                (
                    [slug] =&gt; agents!machine learning
                    [parent] =&gt; agents
                    [name] =&gt; machine learning
                    [canonical] =&gt; 
                    [see] =&gt; 
                    [description] =&gt; 
                )

            [agents!messages] =&gt; Array
                (
                    [slug] =&gt; agents!messages
                    [parent] =&gt; agents
                    [name] =&gt; messages
                    [canonical] =&gt; 
                    [see] =&gt; 
                    [description] =&gt; 
                )

            [agents!methods] =&gt; Array
                (
                    [slug] =&gt; agents!methods
                    [parent] =&gt; agents
                    [name] =&gt; methods
                    [canonical] =&gt; 
                    [see] =&gt; 
                    [description] =&gt; 
                )

        )

)
  --></li>
  <li ><span class="name">AGI</span> - See <span class="has_content"><a href="https://alandix.com/glossary/aibook/artificial general intelligence">artificial general intelligence</a></span><!--  Array
(
    [slug] =&gt; agi
    [parent] =&gt; 
    [name] =&gt; AGI
    [canonical] =&gt; 
    [see] =&gt; artificial general intelligence
    [description] =&gt; 
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/aibook/ai arms race">AI arms race</a></span><!--  Array
(
    [slug] =&gt; ai arms race
    [parent] =&gt; 
    [name] =&gt; AI arms race
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; There is literally an arms race in the military use of AI, but also an AI arms race between commercial companies.   Having the best AI may give massive commercial advantage either in terms of user experience (such as the runaway dominance of Google search due to the {{PageRank}} algorithm) or for behind the scenes activity such as the sue in advertisment placement.
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/aibook/ai winter">AI winter</a></span><!--  Array
(
    [slug] =&gt; ai winter
    [parent] =&gt; 
    [name] =&gt; AI winter
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; The history of AI is often seen in terms of relatve excitement and success followed by periods of disillusionment, called AI winters. The most recent, sometimes simply called 'the AI Winter' was between the mid 1990s, following the early promise and over-hyping of {{neural networks}}  and early 2010s when the devlopemnt of {{deep neural networks}} led to the current high interest in AI.  Some fear that over-inlfalted expections of what AI can achieve may lead to yet another AI winter, but the widespreadsuccessful practical applications of AI makes this seem less likely.
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/aibook/alien intelligence">alien intelligence</a></span><!--  Array
(
    [slug] =&gt; alien intelligence
    [parent] =&gt; 
    [name] =&gt; alien intelligence
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; AI often appears 'intellogent', but in ways thata re very differet to the way a human would tackle the same problem.  One can think of this as 'alien intelligence', misquoting Start Trek, &quot;intelligence, but not as we know it&quot;.  Although light-hearted, this can be a useful way to consider AI in practical applications.
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/aibook/alpha-beta pruning">alpha&ndash;beta pruning</a></span><!--  Array
(
    [slug] =&gt; alpha-beta pruning
    [parent] =&gt; 
    [name] =&gt; alpha&amp;ndash;beta pruning
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; Alpha&amp;ndash;beta pruning is a vaiant of {{minimax search}} for {{zero-sum}} {{two player games}}.  It adopts a similar insight to {{branch and bound}} for simple {{tree search}}.  As there are two players, it keeps a 'best so far' value for each player.  By convention alpha is the best for the first player and beta for the second, but these swop as one moves down the {{game tree}} alernating the players' turns.   Imagine the frist player has exmaind a numer of possible moves and has a best so far move with score 10.  It starts to explore a new move M and looks at the second player's options.  As soon as it encounters one with the second player's score greater than -10 (so first player worse than 10), it can prune the whole of the branch associated with M as it is clearly less good thath best so far move.  A similar strategy is used {{recursively}} but alternating 'best so far' roles.   Note, like {{minimax}}, alpha&amp;ndash;beta pruning assumes a {{perfect opponent}}.
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/aibook/alphafold">AlphaFold</a></span><!--  Array
(
    [slug] =&gt; alphafold
    [parent] =&gt; 
    [name] =&gt; AlphaFold
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; AlphaFold and its successor AlphaFold2 applied some of the technology originally designed for {{AlphaGo}} and applied this to protein folding.  While some of the behaviour of proteins is given by their chemical and molecular structure, the physical shape of a protein, the way it naturally folds given the order of molecular units, is also critical.  For example, this folded shape deterimines which parts of the molecular structure are exposed and which hidden inside and thus the extent to which they are free to interact with other molecules and engage in biological processs.  Determing the folded shape from the chemcal structure is highly complex, and often relied on experimentally observing the structure using electron microscopy and other methods.  AlphaFold revolutionised this field allowing new molecules to be tested for efficacy purely by computer simulation (in silco).  This means only a relatively small number need progress to full lab-based synthesis and testing.
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/aibook/alphago">AlphaGo</a></span><!--  Array
(
    [slug] =&gt; alphago
    [parent] =&gt; 
    [name] =&gt; AlphaGo
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; AlphaGo was one of the key moments of modern AI, when a computer Go player defeated the Go Grandmaster Lee Se-dol.  It had previosuly been believed that the complexity of Go, in particular the large {{branching factor}} of its {{search space}} made it intractable with even near-future technology.  AlphaGo was initially taught using records of large numbers of human games, but then adopted adversarial techniques to improve its performance.  AlphaGo uses traditional computer game techniques, such as {{Monte Carlo tree search}}, guided by heuristics from {{deep neural networks}}.  {{AlphaZero}}, the successor of AlphaGo, dispenses with the need for human training.
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/aibook/alphazero">AlphaZero</a></span><!--  Array
(
    [slug] =&gt; alphazero
    [parent] =&gt; 
    [name] =&gt; AlphaZero
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; AlphaZero is the successor of {{AlphaGo}}, which defeated a Go grandmaster for the first time in 2016.    Whilst {{AlphaGo}} was initialy trained on human games, AlphaZero started from scratch with only a description of the rules of Go and thereafter learning entirely using {{adversarial training}}.
)
  --></li>
  <li ><span class="name"><a href="https://alandix.com/glossary/aibook/ambiguity">ambiguity</a></span>- has subentries<ul class="subentries"><li ><span class="name">in computer vision</span> - See <span class="has_content"><a href="https://alandix.com/glossary/aibook/ambiguous image">ambiguous image</a></span><!--  Array
(
    [slug] =&gt; ambiguity!in computer vision
    [parent] =&gt; ambiguity
    [name] =&gt; in computer vision
    [canonical] =&gt; 
    [see] =&gt; ambiguous image
    [description] =&gt; 
)
  --></li>
<li class="has_content"><span class="name"><a href="https://alandix.com/glossary/aibook/ambiguity!lexical">lexical</a></span><!--  Array
(
    [slug] =&gt; ambiguity!lexical
    [parent] =&gt; ambiguity
    [name] =&gt; lexical
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; {{Ambiguity}} at a lexical level in {{NLP}} is about words, for example &quot;Look at the bow&quot;&quot; might mean look at the front of a ship, at a weapon to shoot arrows or at a person bending before royalty.
)
  --></li>
<li class="has_content"><span class="name"><a href="https://alandix.com/glossary/aibook/ambiguity!pragmatic">pragmatic</a></span><!--  Array
(
    [slug] =&gt; ambiguity!pragmatic
    [parent] =&gt; ambiguity
    [name] =&gt; pragmatic
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; {{Ambiguity}} at a pragmatic level in {{NLP}} is about the meaning of an utterance in context.  For exmaple &quot;it is cold&quot; might be a statement about the temeperature or a request to close the door.
)
  --></li>
<li class="has_content"><span class="name"><a href="https://alandix.com/glossary/aibook/ambiguity!referential">referential</a></span><!--  Array
(
    [slug] =&gt; ambiguity!referential
    [parent] =&gt; ambiguity
    [name] =&gt; referential
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; {{Ambiguity}} at a referential level in {{NLP}} is about waht tjinsg refer to, especially, pronouns.  For example, in &quot;The dog chased the cat. It was fast.&quot; is it the dog or the cat that is fast?
)
  --></li>
<li class="has_content"><span class="name"><a href="https://alandix.com/glossary/aibook/ambiguity!semantic">semantic</a></span><!--  Array
(
    [slug] =&gt; ambiguity!semantic
    [parent] =&gt; ambiguity
    [name] =&gt; semantic
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; {{Ambiguity}} at a semantic level in {{NLP}} is about the meaning of text, for example, &quot;I saw her duck&quot; might I saw her drop down suddenly, or that I saw her pet poultry.  Note here how the zsemantic ambiguity related to {{lexical ambiguity}} of 'her' (as accusative or possessive pronoun) and 'duck' (avoidance action or bird).
)
  --></li>
<li class="has_content"><span class="name"><a href="https://alandix.com/glossary/aibook/ambiguity!syntactic">syntactic</a></span><!--  Array
(
    [slug] =&gt; ambiguity!syntactic
    [parent] =&gt; ambiguity
    [name] =&gt; syntactic
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; {{Ambiguity}} at a syntactic level in {{NLP}} is about structure and {{grammar}}.  For example, does &quot;the man with the dog that is angry&quot; refer to the owner of an angry dog or and angry owner of a dog?
)
  --></li></ul><!--  Array
(
    [slug] =&gt; ambiguity
    [parent] =&gt; 
    [name] =&gt; ambiguity
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; 
    [subentries] =&gt; Array
        (
            [ambiguity!in computer vision] =&gt; Array
                (
                    [slug] =&gt; ambiguity!in computer vision
                    [parent] =&gt; ambiguity
                    [name] =&gt; in computer vision
                    [canonical] =&gt; 
                    [see] =&gt; ambiguous image
                    [description] =&gt; 
                )

            [ambiguity!lexical] =&gt; Array
                (
                    [slug] =&gt; ambiguity!lexical
                    [parent] =&gt; ambiguity
                    [name] =&gt; lexical
                    [canonical] =&gt; 
                    [see] =&gt; 
                    [description] =&gt; {{Ambiguity}} at a lexical level in {{NLP}} is about words, for example &quot;Look at the bow&quot;&quot; might mean look at the front of a ship, at a weapon to shoot arrows or at a person bending before royalty.
                )

            [ambiguity!pragmatic] =&gt; Array
                (
                    [slug] =&gt; ambiguity!pragmatic
                    [parent] =&gt; ambiguity
                    [name] =&gt; pragmatic
                    [canonical] =&gt; 
                    [see] =&gt; 
                    [description] =&gt; {{Ambiguity}} at a pragmatic level in {{NLP}} is about the meaning of an utterance in context.  For exmaple &quot;it is cold&quot; might be a statement about the temeperature or a request to close the door.
                )

            [ambiguity!referential] =&gt; Array
                (
                    [slug] =&gt; ambiguity!referential
                    [parent] =&gt; ambiguity
                    [name] =&gt; referential
                    [canonical] =&gt; 
                    [see] =&gt; 
                    [description] =&gt; {{Ambiguity}} at a referential level in {{NLP}} is about waht tjinsg refer to, especially, pronouns.  For example, in &quot;The dog chased the cat. It was fast.&quot; is it the dog or the cat that is fast?
                )

            [ambiguity!semantic] =&gt; Array
                (
                    [slug] =&gt; ambiguity!semantic
                    [parent] =&gt; ambiguity
                    [name] =&gt; semantic
                    [canonical] =&gt; 
                    [see] =&gt; 
                    [description] =&gt; {{Ambiguity}} at a semantic level in {{NLP}} is about the meaning of text, for example, &quot;I saw her duck&quot; might I saw her drop down suddenly, or that I saw her pet poultry.  Note here how the zsemantic ambiguity related to {{lexical ambiguity}} of 'her' (as accusative or possessive pronoun) and 'duck' (avoidance action or bird).
                )

            [ambiguity!syntactic] =&gt; Array
                (
                    [slug] =&gt; ambiguity!syntactic
                    [parent] =&gt; ambiguity
                    [name] =&gt; syntactic
                    [canonical] =&gt; 
                    [see] =&gt; 
                    [description] =&gt; {{Ambiguity}} at a syntactic level in {{NLP}} is about structure and {{grammar}}.  For example, does &quot;the man with the dog that is angry&quot; refer to the owner of an angry dog or and angry owner of a dog?
                )

        )

)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/aibook/ambiguous image">ambiguous image</a></span><!--  Array
(
    [slug] =&gt; ambiguous image
    [parent] =&gt; 
    [name] =&gt; ambiguous image
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; An ambiguous image is an image that has more than one interpretation.  A classic example is {{Rubin's vase}} which can be seen as a vase or two faces.   Human vision typically flips back and forth, largely unconstrollably, between the interpretations.  In real situatiosn the ambiguity eiether resolves naturally or one moves ones head to geta. different view.  The latter is emulated in {{actve visions}} where the camera moves.
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/aibook/analogy">analogy</a></span>- has subentries<ul class="subentries"><li ><span class="name">derivational</span> - See <span class="has_content"><a href="https://alandix.com/glossary/aibook/derivational analogy">derivational analogy</a></span><!--  Array
(
    [slug] =&gt; analogy!derivational
    [parent] =&gt; analogy
    [name] =&gt; derivational
    [canonical] =&gt; 
    [see] =&gt; derivational analogy
    [description] =&gt; 
)
  --></li>
<li ><span class="name">transformational</span> - See <span class="has_content"><a href="https://alandix.com/glossary/aibook/transformational analogy">transformational analogy</a></span><!--  Array
(
    [slug] =&gt; analogy!transformational
    [parent] =&gt; analogy
    [name] =&gt; transformational
    [canonical] =&gt; 
    [see] =&gt; transformational analogy
    [description] =&gt; 
)
  --></li></ul><!--  Array
(
    [slug] =&gt; analogy
    [parent] =&gt; 
    [name] =&gt; analogy
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; An analogy is when concept, item or situation is compared with another drawing out parallels.  For example, a school teacher with their pupils are like a shepherd herding sheep; in this (implicitluy) the teacher  is in the position of the shepherd and pupils rleated to the sheep.  There are usually limits beyind which the analogy does not hold, for example, one would not expect school childrden to grow wool.  Analogies are used as part of {{human reasoning}} to suggest potential solutions; and forms of this process can be adopted in AI systems  during {{reasoning by analogy}}.   {{Case-based reasoning}} systems may use {{reasoning by analogy}} to transform past cases.
    [subentries] =&gt; Array
        (
            [analogy!derivational] =&gt; Array
                (
                    [slug] =&gt; analogy!derivational
                    [parent] =&gt; analogy
                    [name] =&gt; derivational
                    [canonical] =&gt; 
                    [see] =&gt; derivational analogy
                    [description] =&gt; 
                )

            [analogy!transformational] =&gt; Array
                (
                    [slug] =&gt; analogy!transformational
                    [parent] =&gt; analogy
                    [name] =&gt; transformational
                    [canonical] =&gt; 
                    [see] =&gt; transformational analogy
                    [description] =&gt; 
                )

        )

)
  --></li>
  <li ><span class="name">analytic rules</span> - See <span class="has_content"><a href="https://alandix.com/glossary/aibook/procedural knowledge">procedural knowledge</a></span><!--  Array
(
    [slug] =&gt; analytic rules
    [parent] =&gt; 
    [name] =&gt; analytic rules
    [canonical] =&gt; 
    [see] =&gt; procedural knowledge
    [description] =&gt; 
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/aibook/analytical engine">Analytical Engine</a></span><!--  Array
(
    [slug] =&gt; analytical engine
    [parent] =&gt; 
    [name] =&gt; Analytical Engine
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; The Analytical Engine was a design by {{Charles Babbage}} in the 1830s following his {{Difference Engine}}, inspired by the punch-card-based  {{Jacquard Looms}}    It was never completed, but in concept is seen as the first programable computer.  {{Ada Lovelace}}, long term collaborator with Babbage idetified the special qualities of the ``{\em The Difference Engine can merely tabulate, and is incapable of developing, the Analytical Engine can either tabulate or develope.}''
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/aibook/anderson, john robert">Anderson, John Robert</a></span><!--  Array
(
    [slug] =&gt; anderson, john robert
    [parent] =&gt; 
    [name] =&gt; Anderson, John Robert
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; Anderson developed the highly influential {{ACT*}} cognitive architecture.
)
  --></li>
  <li ><span class="name">Andrey Andreyevich Markov</span> - See <span class="has_content"><a href="https://alandix.com/glossary/aibook/markov, andrey">Markov, Andrey</a></span><!--  Array
(
    [slug] =&gt; andrey andreyevich markov
    [parent] =&gt; 
    [name] =&gt; Andrey Andreyevich Markov
    [canonical] =&gt; 
    [see] =&gt; markov, andrey
    [description] =&gt; 
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/aibook/anonymisation">anonymisation</a></span><!--  Array
(
    [slug] =&gt; anonymisation
    [parent] =&gt; 
    [name] =&gt; anonymisation
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; General Data Protection Regulation (GDPR) is the European Union {{data protection law}} intended, amongst other things, to protect the {{privacy}} of EU citizens.  It is regarded as one of the strongest {{data protection laws}} in the world and has international significane.  This is not just because the EU itself is a large market, but because it applied extra-terrotorially if internet or web-based systems are available for use in the EU.  This is of particular relevance in AI settings as it can be hard to be sure what information is preserved in the weights of a {{deep neural networks}}, emphasisng the importance of {{explainable AI}}. 
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/aibook/anonymous identifier">anonymous identifier</a></span><!--  Array
(
    [slug] =&gt; anonymous identifier
    [parent] =&gt; 
    [name] =&gt; anonymous identifier
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; An anonymous identifier is a number or string that is unique to the person, but from which it is hard or impossibel to determine which person it refers to.  It is used, in {{pseudonymised}} data to help connect together different record that refer to the same indivdiual.  Sometiems, the lost that connects individuals with identifiers is deleted after use in order to prevent accidental disclosure, but sometime sit has to be retaianied in order to correctly pseudonymise further data.  In these case the {{authority file}} linking ids to people needs to be maintained at a much higher level of security to the {{pseudonomised data}} itself.
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/aibook/anti-discrimination laws">anti-discrimination laws</a></span><!--  Array
(
    [slug] =&gt; anti-discrimination laws
    [parent] =&gt; 
    [name] =&gt; anti-discrimination laws
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; Many nations have laws to prohibit particular forms of descriminiation often including gender, ethnicity and sexual orientation.  For example, the UK list of {{protected characteristics}} at the time of writing includes  age, gender reassignment, being married or in a civil partnership, being pregnant or on maternity leave, disability, race including colour, nationality, ethnic or national origin, religion or belief, sex, and sexual orientation.  The laws may have severe financial or other penalties for companies who have been found to have acted in a discrimnatory manner, for example, having software that has some form of {{bias}}.   This is parricularly probematic for compex AI algorithms and machine learning such as {{deep neural networks}} as it may be unclear what aspects are being used to make a decision.  Note that irrespectuve of the law in the nation(s) in which you are working or deploying systems, you should consider the {{ethical}} consequences of systems you produce, for example, bias for non-protected characteristcs such  social background, or wider {{societal impact of AI}}.
)
  --></li>
  <li ><span class="name">antecedent-driven reasoning</span> - See <span ><a href="https://alandix.com/glossary/aibook/bottom-up reasoning">bottom-up reasoning</a></span><!--  Array
(
    [slug] =&gt; antecedent-driven reasoning
    [parent] =&gt; 
    [name] =&gt; antecedent-driven reasoning
    [canonical] =&gt; 
    [see] =&gt; bottom-up reasoning
    [description] =&gt; 
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/aibook/apache hadoop">Apache Hadoop</a></span><!--  Array
(
    [slug] =&gt; apache hadoop
    [parent] =&gt; 
    [name] =&gt; Apache Hadoop
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; Hadoop is the Apache {{distributed computation}} system to allow efficient large scale processing of big data.  It includes an open source implementation of {{MapReduce}}.
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/aibook/aperture">aperture</a></span><!--  Array
(
    [slug] =&gt; aperture
    [parent] =&gt; 
    [name] =&gt; aperture
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; The aperture of a camera lens is the size of the hole through whuch light enters the camera.  This controls the amount of light entering the camera, so has to be larger in darker conditions.  It also affects the {{depth of field}} (what is in focus simultaneously) as the outer areas of the lens have more distortion, so that a smaller apperture leads to sharper images even if they are not at the exact {{focal point}} ... and also, becasue your pupil acts as apperture control in your eye, the reason why it is easier to read in bright light.
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/aibook/appropriate intelligence">appropriate intelligence</a></span><!--  Array
(
    [slug] =&gt; appropriate intelligence
    [parent] =&gt; 
    [name] =&gt; appropriate intelligence
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; Appropriate intelligence is a deisgn heuristic for {{intelligent user interfaces}}.   When creating an ntelligent interface the temptation is to focus on making the AI as good as possible, but it is usually more importnt to make sure that the inetraction deisgned aorund the AI makes an overall system that works well.  In particular, it is important that when the AI fails (which it will!) this does not overly inconvenience the user.  Positive examples incude the Microsoft SUM (&amp;Sigma;) button and the design of {{onCue}}, whereas {{Clippy}} is an example of poor use of intelligence. as it interupted the user's workflow.
)
  --></li>
  <li ><span class="name">appropriating technology</span> - See <span class="has_content"><a href="https://alandix.com/glossary/aibook/appropriation">appropriation</a></span><!--  Array
(
    [slug] =&gt; appropriating technology
    [parent] =&gt; 
    [name] =&gt; appropriating technology
    [canonical] =&gt; 
    [see] =&gt; appropriation
    [description] =&gt; 
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/aibook/appropriation">appropriation</a></span><!--  Array
(
    [slug] =&gt; appropriation
    [parent] =&gt; 
    [name] =&gt; appropriation
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; We often use objects and tools for purposes far different than their intended use, for example, cutting parcel tape with a key or proping open a door with a heavy book.  This process of appropriation can be used for technology, for example colouring cells in a spreadsheet to act as a rough image.
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/aibook/architecture">architecture</a></span><!--  Array
(
    [slug] =&gt; architecture
    [parent] =&gt; 
    [name] =&gt; architecture
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; Within computing architecture usually refers to the way in which software is structured.  Within AI this may include {{blackboard architectures}} and the choice of the types and sizes of the different layers in a {{deep neural network}}.  In addition a {{cognitive arcitecture}} is a way of representing the main processes and units in the human (or other animal) brain, but at a much higher level than individual {{neorones}}.
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/aibook/area framework">AREA framework</a></span><!--  Array
(
    [slug] =&gt; area framework
    [parent] =&gt; 
    [name] =&gt; AREA framework
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; The AREA framework hase been developed by EPSRC, the UK funding council that covers AI and computing, in order to promote {{responsible innovation}}.  The framework encourages researchers to consiser theor work under four headings: Anticipate, Reflect, Engage, and Act, each of which has further levels of advice.
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/aibook/area under the curve">area under the curve</a></span><!--  Array
(
    [slug] =&gt; area under the curve
    [parent] =&gt; 
    [name] =&gt; area under the curve
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; The area under the curve is an {{accuracy}} measure and is literally the area under the {{ROC}} curve.  If a decision rule has higher {{precision}} or higher {{recall}} it ends up with a larger area under the curve.  A {{random classifer}} has an area under the curve of 0.5, wheras a {{perfect classifier}} with 100% {{precsion}} and 100% {{recall}} has an area undet the curve of 1.  In general, because of the {{precision-recall trade-off}} the value lies somewhere between the two.
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/aibook/arithmetic mean">arithmetic mean</a></span><!--  Array
(
    [slug] =&gt; arithmetic mean
    [parent] =&gt; 
    [name] =&gt; arithmetic mean
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; The arithmetic mean (often simply the mean) adds up the values of a numercal feature ina dataset, for example, the hights of people ina. room, and divides by the total number of indviduals.  For example, the mean of the numbers [1,7,6,2,4] is (1+7+6+2+4)/5 = 20/5 = 4.  The term {{average}} may refer to the arithmetic mean or may refer to the {{median}}.  The {{median}} is often a better {{centrality}} measure than the mean as it is more stable espedcially given extreme values, or for {{non-Normal distributions}} such as the {{power law distributions}}.  However, the mean has nice mathematical properties that make it easier to process, for example, if you have caclualted the mean of two datasets (say m&lt;sub&gt;1&lt;/sub&gt; and m&lt;sub&gt;2&lt;/sub&gt;) and know how many in each  (say N&lt;sub&gt;1&lt;/sub&gt; and N&lt;sub&gt;2&lt;/sub&gt;) it is possible to combine these: &lt;br&gt; overall mean = (m&lt;sub&gt;1&lt;/sub&gt; x N&lt;sub&gt;1&lt;/sub&gt; + m&lt;sub&gt;2&lt;/sub&gt; x N&lt;sub&gt;2&lt;/sub&gt;) / (N&lt;sub&gt;1&lt;/sub&gt; + N&lt;sub&gt;2&lt;/sub&gt;)
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/aibook/arma">ARMA (Auto Regressive Moving Average)</a></span><!--  Array
(
    [slug] =&gt; arma
    [parent] =&gt; 
    [name] =&gt; ARMA (Auto Regressive Moving Average)
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; ARMA models use a combination of {{auto-regressive}} and {{moving average}} techniques to model time series.  It is a form of {{linear model}} for stochastic processes.
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/aibook/artificial general intelligence">artificial general intelligence</a></span><!--  Array
(
    [slug] =&gt; artificial general intelligence
    [parent] =&gt; 
    [name] =&gt; artificial general intelligence
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; Artificial general intelligence (AGI) is the idea or goal of proecuing AI that can be applied across all, or at least a wide range, of domains and problems.  Early AI systems tended to be very specific in the kinds of problems they addressed based partly on their mode of {{reasoning}} and partly on their {{knowledge base}}.  The use of {{deep leaerning}} and in particular {{large-language models}} has rasied the possibility that these systems, currently relying on large volumes of himan generated materials, may already be at the edge of AGI.   Others disagree, especially as these systems are effectively reaching the limits of human-generated data and will have to rely on some form of {[self learning}}.
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/aibook/artificial intelligence">artificial intelligence</a></span><!--  Array
(
    [slug] =&gt; artificial intelligence
    [parent] =&gt; 
    [name] =&gt; artificial intelligence
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; Artificial intelligence has many definitions, but is broadly about computers that do things which, if performed by a human, would be considered intelligent.   Proponents of {{strong AI}} believe that it  will ultimately be possible to create machine intelligence that is equivalent in all sense to a human; whereas others take a {{weak AI}} approach that is happy to see inspiration from human intelligence realised in machines.  A more pragmatic approach simply looks for algorithms and techniques that can be used in practical applications.
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/aibook/artificial life">artificial life</a></span><!--  Array
(
    [slug] =&gt; artificial life
    [parent] =&gt; 
    [name] =&gt; artificial life
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; The term artificial life refers to a range of types of model, from those that operate are a level closer to baterium to human socities.  A common low-level exmaple are {{celluar automota}} such as  {{Conway}}'s {{Game of Life}}.  Higher level examples include models of {{flocking behaviour}} or modelling of human movements in emergency planning.  {{Swarm computing}} can also be seen as a form of artificial life. &lt;br&gt;&lt;br&gt; Artificial life can alos refer to the design of biological matter, often using AI to drive the process.  This includes {{xenobots}}, groups of cells desitned to act like living robots; and even completely designing DNA from scratch.
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/aibook/artificial neural networks">artificial neural networks</a></span><!--  Array
(
    [slug] =&gt; artificial neural networks
    [parent] =&gt; 
    [name] =&gt; artificial neural networks
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; Artificial neural networks is another term for computer-based {{neural networks}}, largely used when one wants to distinguish them from biological neural networks in the brain.  This is most likely to be the case in more fundamental work where artficial networks are being used in neuroscience to understand brain behaviour or where novel algorithms are being developed based on brain neural analogies.  In areas where the digital nature is clear the term {{neural network}} is used.
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/aibook/artificial society">artificial society</a></span><!--  Array
(
    [slug] =&gt; artificial society
    [parent] =&gt; 
    [name] =&gt; artificial society
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; Artificial societies are {{agent-based models}} to emulate social phenomena, current and hsitorical, for example neolithic migration patterns.  Relatively simple models of individuals often give rise to overall behaviours that emulate actual social processes, rather like {{flocking modles}} do for animal movements.
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/aibook/asimov, isaac">Asimov, Isaac</a></span><!--  Array
(
    [slug] =&gt; asimov, isaac
    [parent] =&gt; 
    [name] =&gt; Asimov, Isaac
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; Science fiction author famous for his {{three laws of robotics}}.
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/aibook/aspect ratio">aspect ratio</a></span><!--  Array
(
    [slug] =&gt; aspect ratio
    [parent] =&gt; 
    [name] =&gt; aspect ratio
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; For images such as photographs, this is the ratio between the width and height of the image.  For example a square image has an aspect artion of 1:1 whereas a traditional photograph is 4:3, that is a 1/3 wider than it is high.   For a digital image the aspect ratio usually refers to the ratio of the number of pixels, for example a 1024x768 has a 4:3 aspect ration, wheas a wide screen at 1920x1080 has an aspect ration of 16:9.  Most commoly each pixel is square, and this is certainly the case on almost all computer screens, but occasionally in specialised applications pixels may be rectangular an have an asoect ration that is not 1:1; when this is the case the apect artio of the overall inage and the pixel rations wouydnot be the same, for example a 1000x500 image where each pixel had a 3:2 aspect ration would have an overall aspect ratio of 3:1 when printed/viewed.
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/aibook/associative memory">associative memory</a></span><!--  Array
(
    [slug] =&gt; associative memory
    [parent] =&gt; 
    [name] =&gt; associative memory
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; An assocative memory is one which tries to learn relationshop between an imput and an output so that when given an unseen input it can allocate an appropriate output.  In some case the output is very different form the output. For example, where an image is associated with a class.  In others the memory can be used as an {{auto-encoder}} where the input and output are identical -- this form of {{auto-associative memory}} initially does not seem useful, however it may be used where the input is in some way incomplete or noisy and the memory recreates a full or noise-free image.
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/aibook/asymmetric distribution">asymmetric distribution</a></span><!--  Array
(
    [slug] =&gt; asymmetric distribution
    [parent] =&gt; 
    [name] =&gt; asymmetric distribution
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; An asymmetric distribution is a {{probability distrubution}} that is asymetric about its {{median}}. For example, waiting times in many contexts are often highly asymetric, not least because ther cannot be negative waiting times, but aso becasue there may also be occasional long waiting times.   Asymmetric distributions are often heavily {{skewed}}, as in the example of waiting times, with a {{long tail}} on one side, but may also simply be different shapes.
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/aibook/attention mechanisms">attention mechanisms</a></span><!--  Array
(
    [slug] =&gt; attention mechanisms
    [parent] =&gt; 
    [name] =&gt; attention mechanisms
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; Attention mechanisms are used in {{machine learning}}, particularly for text and in {{large language models}} and other forms of {{deep learning}}.  The concept borrows form human attention, which is critical for cognition.  Given a sequence of tokens (e.g. words in a text), simple {{window-based methods}} would treat all past tokens in the window as equally relevant for predicting or otherwise leaening based on a current token.  In contrast attention mechanisms attempt to work out past tokens that are especially relevant and gves these higher weight (attention) during training.  This may be acheived by creating an interest vector for each input (as part of the {{deep learning}} process) that in some way represents its topic area and then matching the interest vector of past tokens with the current token.  Some attention mechanisms will also retain past tokens that have proved especially relevant based on their relevance to later tokens, thus allowing memory beyond the window.  {{Salience}} can be combined with attention mechanisms by rating the unusuallness of tokens and token sub-seqences and use these to weight tokens together with similarity to the current token.
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/aibook/augmented transition network">augmented transition network</a></span><!--  Array
(
    [slug] =&gt; augmented transition network
    [parent] =&gt; 
    [name] =&gt; augmented transition network
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; An augmented transition network is a {{transition network}} that allows procedures to be attached to arcs that act as an additional test to allow the arc to be traversed.  In particular this offers a mechanism to model {{context sensitive grammars}} and hence more complex patterns of input than would be possible with a simple {{transition network}}.
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/aibook/authority file">authority file</a></span><!--  Array
(
    [slug] =&gt; authority file
    [parent] =&gt; 
    [name] =&gt; authority file
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; An authority file is used to ensure that named people, places or other things can be connected in a dataset.  For example, the name 'Alan Dix' might refer to the author of AI, HCI and statistics textbooks, or Alan Dix the theatre maker.  An authrity file would allocate each a unique identifier, say AlanDix_276493 and AlanDix_835672 and then these identifiers would be used in orther data records to ensure it is clear whcih Alan Dix is being fereed to.  In {{pseudonomised data}} the identifeir is deliberately chosen to be anonymous (unlike those in the example) and the authorty file is either kept securely or deleted after use.
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/aibook/auto-associative memory">auto-associative memory</a></span><!--  Array
(
    [slug] =&gt; auto-associative memory
    [parent] =&gt; 
    [name] =&gt; auto-associative memory
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; An auto-associative memory is an {{assocative memory}} where the input and output are identical.  This initially does not seem useful, however it may be used where the input is in some way incomplete or noisy and the {{auto-encoder}} recreates a full or noise-free image.
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/aibook/auto-regressive model">auto-regressive model</a></span><!--  Array
(
    [slug] =&gt; auto-regressive model
    [parent] =&gt; 
    [name] =&gt; auto-regressive model
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; An auto-regressive model (of order n) is used when a time series is believed to be of the form x&lt;sub&gt;t&lt;/sub&gt; =  a&lt;sub&gt;1&lt;/sub&gt; x&lt;sub&gt;t-1&lt;/sub&gt; + &hellip; + a&lt;sub&gt;1&lt;/sub&gt; x&lt;sub&gt;t-(n-1)&lt;/sub&gt; + noise&lt;sub&gt;t&lt;/sub&gt;.  The simplest example is a {{drunkards walk}} where n=1 and a&lt;sub&gt;1&lt;/sub&gt;=1.
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/aibook/autoencoder">autoencoder</a></span><!--  Array
(
    [slug] =&gt; autoencoder
    [parent] =&gt; 
    [name] =&gt; autoencoder
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; Autoencoder is another word for {{autoassocitve memory}}, but more often used in recent {{neural network}} literature.  In the training data the input and output are identical, often the same image as each other.   The network thius learns to associate the image with itself.  This initially does not seem very intersting except that internal layers typically are substantially smaller than the full input.  A small middle layer is effectvely coding the key features of the image for reconstruction.  Becasue of this the inner, {{pinch point}} layer can sometimes be used as a form of {{data-reduction}} for other knds of network, for example a classifier.  Alternatively the network may be able to recreate images from partial inputs or even generate new ones by setting the inner layer.
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/aibook/automated decision">automated decision</a></span><!--  Array
(
    [slug] =&gt; automated decision
    [parent] =&gt; 
    [name] =&gt; automated decision
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; An automated decsion is one where the final decsion is made by the computer system without a human having the final say.  For sensitive decsions such as medical or military applications, many prefer there to be a {{human in the loop}} at very least to veto or put a check on the automated decisions.  Humans are almost always still involved in fully automated, either in supplying data or executing the actions.  For example, in applications such as Uber or Lyft the descison about which car is sent to which client is completely automated even though humans drive the cars.
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/aibook/automation bias">automation bias</a></span><!--  Array
(
    [slug] =&gt; automation bias
    [parent] =&gt; 
    [name] =&gt; automation bias
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; Automation bias is the tendency for humans to agree with computer advice or be swayed by computational data even when the automated system.  This can be problematic when the system fails, is error prone or doesn't take into account all aspects of a situation.   For example, an AI system may use a small number of features to advise on whether someone is at high or low risk when they enter an hospital emergnecy department; if the person is obviously not breathing the paramedic or nurse shoudl clearly igbore this, but there can be a tendency to assume the computer knows best.  This can also lead to {{bias}} for example, the {COMPAS}} system that has been used by judges as part of parole decisions in the US has been criticsed as embodying elements of racial bias.  ULtimately the judge makes the decsion, but may be swayed by the system's assessment.
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/aibook/autonomous vehicle">autonomous vehicle</a></span><!--  Array
(
    [slug] =&gt; autonomous vehicle
    [parent] =&gt; 
    [name] =&gt; autonomous vehicle
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; An autonomous vehicle is one that partially or fully drives itself.   In a semi-autonmous mode, this may aid particular manouvres such as parallel parking, or driving within lane markings on a long drive.  In fully autonomous mode, the car also has to deal with ever changing road conditons; other road users including pedestrians and other vehicles; and hazards suhc aas child or animal running onto the road. This usually involves a combination of vision and other sensors linked to rapid response planning and hazard avoidance.   
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/aibook/autonomous weapons">autonomous weapons</a></span><!--  Array
(
    [slug] =&gt; autonomous weapons
    [parent] =&gt; 
    [name] =&gt; autonomous weapons
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; An autonomous weapon is any form of missile, drone or armed vehicle that uses AI to make its own decsions.  This may be about navigating terrain, but may also include choosing targets.  The latter has given rise to substantial concern as there is no {{human in the loop}} to  ensure that only acceptable military targets are attacked.  AI is used extensively in other military applications, including offering advice on targetting, but without the systems being completely autonomous.
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/aibook/astar algorithm">A<sup>*</sup> algorithm</a></span><!--  Array
(
    [slug] =&gt; astar algorithm
    [parent] =&gt; 
    [name] =&gt; A&lt;sup&gt;*&lt;/sup&gt; algorithm
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; The A&lt;sup&gt;*&lt;/sup&gt; algorithm is a {{heuristic search}} most often used for finding shortest-path routes.  If the {{heuristic}} guarantees a {{lower bound}} on the costs of any node below, then this can be used to {{prune}} whole branches where the heurisic exceeds the current best solution.  That is, it can be seen as heuristic version of {{branch and bound}}.  Often the raw {{heauristic}} is about the path beyond the node, for example the crows-fly distance to the target location in geographic {{shortest path}} search, in which case the distance along the path to the node has to be added to the raw heuristic to create a total path {{lower bound}}.  The heauristic can also be used to guide the next node to explore in a variant of {{hill climbing}}.
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/aibook/babbage-charles">Babbage, Charles</a></span><!--  Array
(
    [slug] =&gt; babbage-charles
    [parent] =&gt; 
    [name] =&gt; Babbage, Charles
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; Charles Babbage was an nineteenth century inventor, famous for his {{Difference Engine}}, which used mechanical means to compute complex calculations, and is thus seen as one of the pioneers of computing.  He later invdented a more complex programmable device, the {{Analytic Engine}}, which is regarded as the first programmable computer, with his collaborator, {{Ada Lovelace}}, as the first programmer.
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/aibook/backpropagation">backpropagation</a></span><!--  Array
(
    [slug] =&gt; backpropagation
    [parent] =&gt; 
    [name] =&gt; backpropagation
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; Backpropogation is a {{supervised learning}}  for multi-level {{neural networks}}.   For each training example, it takes the difference between the expected and actual output at the final layer and then uses the {{differential}} of the {sigmoid}}  function at each node to work out error values at earlier layers and hence update the weights on links between nodes.
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/aibook/backtracking">backtracking</a></span><!--  Array
(
    [slug] =&gt; backtracking
    [parent] =&gt; 
    [name] =&gt; backtracking
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; Backtracking is when a {{search}} or related algorithm returns to previous state.  This maybe because the algorithm has got stuck, for example when a {{tree search}} encounters a node that is {{infeasible}}, that is does not satisfy all of the {{constraints}} for an acceptable solution.  Alternatively, when the algorithm has found some form of a soluution, it may want to look for others.
)
  --></li>
  <li ><span class="name">backward branching factor</span> - See <span class="has_content"><a href="https://alandix.com/glossary/aibook/branching factor">branching factor</a></span><!--  Array
(
    [slug] =&gt; backward branching factor
    [parent] =&gt; 
    [name] =&gt; backward branching factor
    [canonical] =&gt; 
    [see] =&gt; branching factor
    [description] =&gt; 
)
  --></li>
  <li ><span class="name">backward chaining</span> - See <span class="has_content"><a href="https://alandix.com/glossary/aibook/backward reasoning">backward reasoning</a></span><!--  Array
(
    [slug] =&gt; backward chaining
    [parent] =&gt; 
    [name] =&gt; backward chaining
    [canonical] =&gt; 
    [see] =&gt; backward reasoning
    [description] =&gt; 
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/aibook/backward reasoning">backward reasoning</a></span><!--  Array
(
    [slug] =&gt; backward reasoning
    [parent] =&gt; 
    [name] =&gt; backward reasoning
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; Backward reasoning is used in {{reasoning systems}} when rules are run'backwards' in order to ask &quot;what would need to be true in order to show X&quot;. That is, given a goal to find sub-goals that need to be solved, themsleves using backward reasoning.  It is particularly useful when the backward {{branching factor}} is smaller than the forwards {{branching factor}}, for example wgere there are huge numbers of fact known combing them to generate derived knowledge through {{forward reasoning}} would generate vast numbers of anticident facts in the hope that one of them iis useful.  In contrast only a small number of them will be relevant to particular goal.  In an {{expert system}} backward reasoning can be used to suggest questiosn to ask; for example, given &quot;all men are mortal&quot;, and the goal is to prove &quot;Socrates is mortal&quot;, a backward reasoning system might ask &quot;Is Socrates a man?&quot;
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/aibook/bag of words">bag of words</a></span><!--  Array
(
    [slug] =&gt; bag of words
    [parent] =&gt; 
    [name] =&gt; bag of words
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; A document consists of structured text: sentances, paragaphs, and meaningful phrases.  Bag of words techniques ignore this structure reducing the document to the set of words with a frequency count for each.  This is used for {{similarity metrics}} such as the {{Jaccard similarity}} and {{cosine similarity}}.
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/aibook/bard">Bard</a></span><!--  Array
(
    [slug] =&gt; bard
    [parent] =&gt; 
    [name] =&gt; Bard
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; Bard is a {{Google}} {{chatbot}} based on a {{large language model}}, a competitor to ChatGPT.
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/aibook/base rate">base rate</a></span><!--  Array
(
    [slug] =&gt; base rate
    [parent] =&gt; 
    [name] =&gt; base rate
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; The base rate is the underlying frequency or proability of a phenomena.  For example, if one in thousand people are likely to be suffering from flu during the windter, then the base rate of flu is 0.001.  The base rate is critical when using evidence to make diagnoses or ither decisions.  For example, if flu and the common cold have similar symptoms and ther base rate of the common cold is one in 50, then someone with a runny nose and headache is more likely to be suffering form a cold than flu.  This is captured formally in {{Bayes theorem}}.  &lt;br&gt;&lt;.br&gt;Base rates are also important when trying to reduce {{bias}} or {{discrimination}} as we try to be fair to an individual irrespectve of the base rate for their demographic.  Without taking remedial action data-based {{machine learning}} will tend to be discriminatory becasue of this, even if the data is free of human bias.
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/aibook/bayes theorem">Bayes Theorem</a></span><!--  Array
(
    [slug] =&gt; bayes theorem
    [parent] =&gt; 
    [name] =&gt; Bayes Theorem
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; Bayes theorem is a part of {{probability}} theory that relates conditional probabilities and can be used to update beliefs (expressed as probabilities) to be updated based on new evidence.  This is used in a number of approaches in AI including {{Bayesian networks}}.  To be precise Bayes theorem says: &lt;br&gt;. P( X | E ) = ( P( E | X ) P(X) ) / sum ( P( E | X&lt;sub&gt;i&lt;/sub&gt; ) P(X&lt;sub&gt;i&lt;/sub&gt;) ) &lt;br&gt; where X is some unknown state (such as &quot;it is raining&quot;) the sum is takend over all possible states (e.g. &quot;it is dry&quot;, &quot;it is snowing&quot;) and E is some other observable (often interpreted as evidence, such as &quot;there's a wet umbrella in the hall-stand&quot;).  It effectively allows one to turn condirtonal probabiliies round.  When being used to update beliefs using evidence the probabilities  P(X&lt;sub&gt;i&lt;/sub&gt;) are known as the {{prior distribution}}, whereas P(X&lt;sub&gt;j&lt;/sub&gt;|E) is the {{posterior distribution}.
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/aibook/bayesian methods">Bayesian methods</a></span><!--  Array
(
    [slug] =&gt; bayesian methods
    [parent] =&gt; 
    [name] =&gt; Bayesian methods
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; This refers to any kinds of methods that use {{Bayes theorem}}.
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/aibook/bayesian network">Bayesian network</a></span><!--  Array
(
    [slug] =&gt; bayesian network
    [parent] =&gt; 
    [name] =&gt; Bayesian network
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; A Bayesian network is a {{network}} of evidence and deriveed properties/facts linked by {{conditional probabilities}} that can be used to work out the probabilities of various facts given evidence. The same network can typically be used to make inferences in different dircetions, for example working out the probability of seeing someone with an umbrella if it is rainy, or the probability that it is rainy if you spot someone with an umbrella.
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/aibook/berners-lee, tim">Berners-Lee, Tim</a></span><!--  Array
(
    [slug] =&gt; berners-lee, tim
    [parent] =&gt; 
    [name] =&gt; Berners-Lee, Tim
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; Tim Berners-Lee is the inventor of the {{World-Wide Web}}, which he developed while working at {{CERN}} initially as a means of sharing scientific data.
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/aibook/best first search">best first search</a></span><!--  Array
(
    [slug] =&gt; best first search
    [parent] =&gt; 
    [name] =&gt; best first search
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; Best-first search is a form of {{tree search}} where there is a {{heuristic evaluation}} for each node. At each step, the best-first algorithm chooses the node with the best heuristic to follow down the tree, only {{backtracking}} when it gets to a leaf node, or in some other way reaches em impasse.
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/aibook/bias">bias</a></span><!--  Array
(
    [slug] =&gt; bias
    [parent] =&gt; 
    [name] =&gt; bias
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; A system has bias if it unfairly benfits or is more positive to one gorup of people compared with another.   Bias can arise for many reasons (see {{chap:bias}}) including when the training data in some way embodies human discrimination; this may happen even if the algorithm appears to be being fair.  Bias is diffidult to avoid entirely as it depends on what we consider fair characetrstics, for example, job recruitment might quite reasonably use exam results of relevant subjects, so is in some ways biased against those who, for no fault of their own , have not had as good an education.  What is deemed acceptable and avodiing unacceptable bias is in part a personal and corporate ethical decsion, but is also a legal one as many nations have {{anti-discrimination laws}}.    Note the term bias also has a very precise statsitsical meaning relating to the whether or not the long-term output of a statistical measure of some phenomenon is equal to the true value.
)
  --></li>
  <li ><span class="name"><a href="https://alandix.com/glossary/aibook/bibliographic database">bibliographic database</a></span><!--  Array
(
    [slug] =&gt; bibliographic database
    [parent] =&gt; 
    [name] =&gt; bibliographic database
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; 
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/aibook/big blue">Big Blue</a></span><!--  Array
(
    [slug] =&gt; big blue
    [parent] =&gt; 
    [name] =&gt; Big Blue
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; IBM Big Blue is the chess program that defeated the grandmaster Gary Kasprov in 1997.  This was a shock at the time as chess at grandmaster was considered too difficult for a computer.  Deep Blue made extensive use of 'brute force'  {{lookahead}} running on an IBM supercomputer and Kasprov labelled it an 'alien opponent' as it played in a very different way to a human.
)
  --></li>
  <li ><span class="name"><a href="https://alandix.com/glossary/aibook/big data">big data</a></span><!--  Array
(
    [slug] =&gt; big data
    [parent] =&gt; 
    [name] =&gt; big data
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; 
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/aibook/binary image">binary image</a></span><!--  Array
(
    [slug] =&gt; binary image
    [parent] =&gt; 
    [name] =&gt; binary image
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; A binary image is a black-and-white {{image}} where the {{pixels}} only take the values 1 or 0.
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/aibook/bitmap image">bitmap image</a></span><!--  Array
(
    [slug] =&gt; bitmap image
    [parent] =&gt; 
    [name] =&gt; bitmap image
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; A bitmap image represents a visual scene or {{image}} as a matrix of {{pixesl}} each of which is a binary representation of a colour.  This may be direct, for example encoding red, blue and green levels each as 8 bit values, or may be indexed, wheer each pixel is an index into a colour map.  The latter is less common in modern formats.  Often bitmap images are {{compressed}} so that the storage format does not driectly represent the {{pixels}}, but these can be restored for display or processing.
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/aibook/blackboard architecture">blackboard architecture</a></span><!--  Array
(
    [slug] =&gt; blackboard architecture
    [parent] =&gt; 
    [name] =&gt; blackboard architecture
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; Blackboard architectures are used in {{agent}}-based AI systems to share information between {{agents}}.  the blackboard is a gared information structure avaiable to all the agents.  Each agent mnitors the blackboard and when it spots soemthing ot knows how towrk with takes the elemnt, processs it and then posts is results back onto the blackbiard in the hope that anoter agent will pick it up.   &lt;br&gt;&lt;br&gt;As an example, consider a maths solving AI system.  There is one agent (call it A) that knows about basic adding up, another (call it B) that knows how group things into brackets, one (call it S) that can swop sides, M that ca multiple/divide sides by the sam thing, and others still for other kinds of manipulation. Initially the blackbaord has the equation &quot;3x = x+6&quot;.  When agent A looks at it, it can only do number arithmetic, similarly B can do nothing, but agent S can process it and posst back &quot;3x &ndash; x = 6&quot;.  At this point B kicks in and posts back &quot;(3&ndash;1)x = 6&quot;.  A now has some basic aritjmatic so posts back &quot;2x = 6&quot;, M sees it and posts &quot;x = 6/2&quot;, finally agent A posts &quot;x = 3&quot;, the solution.&lt;br&gt;&lt;br&gt;In practice agent archotectures will have various machnaicms to make this process more efficient such as ways for an agent to register patterns that are of interest to avoid constant {{polling}} of the blackboard; there may also be some sort of coordinating agent, that helps to prioritise the agents.  
)
  --></li>
  <li ><span class="name"><a href="https://alandix.com/glossary/aibook/boden-margaret">Boden, Margaret</a></span><!--  Array
(
    [slug] =&gt; boden-margaret
    [parent] =&gt; 
    [name] =&gt; Boden, Margaret
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; 
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/aibook/boltzmann machine">Boltzmann machine</a></span><!--  Array
(
    [slug] =&gt; boltzmann machine
    [parent] =&gt; 
    [name] =&gt; Boltzmann machine
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; The Boltzmann machine is an early form of {{neural network}} based on {{spin glass models}} in physics, which also have parallels in human brain mechanisms.   Boltzman machines are often used as an {{autoencoder}}.  Boltzman machines have two types of nodes (or layers): visible (which can be used for input and output) and hidden (which often act as low-demiensional encodings of inputs).  In a standard Boltzmann machine all nodes are fully interconnected, but in {{restricted Boltzmann machines}} there are only connections between the layers.
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/aibook/bombe">Bombe</a></span><!--  Array
(
    [slug] =&gt; bombe
    [parent] =&gt; 
    [name] =&gt; Bombe
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; The Bombe was a computational device designed by {{Alan Turing}} and others at Bletchley to help decrypt {{Enigma machine}} messages.  This was one of the inspiratons for his later work studying computation.
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/aibook/boolean network">Boolean network</a></span><!--  Array
(
    [slug] =&gt; boolean network
    [parent] =&gt; 
    [name] =&gt; Boolean network
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; A Boolean network is a form of neural netwrok where all of the values are Boolean (true/false) and the internal nodes logic functions such as AND or NOR.  It is can be hard to train a Boolean network as they are discontinuous, but they can be very efficient to run, especially if compiled into a {{FPGA}}.  In some case more compelx {{neural netowrks}} can be simplified into Boolean networks.
)
  --></li>
  <li ><span class="name"><a href="https://alandix.com/glossary/aibook/boosting">boosting</a></span><!--  Array
(
    [slug] =&gt; boosting
    [parent] =&gt; 
    [name] =&gt; boosting
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; 
)
  --></li>
  <li ><span class="name"><a href="https://alandix.com/glossary/aibook/bootstrapping">bootstrapping</a></span><!--  Array
(
    [slug] =&gt; bootstrapping
    [parent] =&gt; 
    [name] =&gt; bootstrapping
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; 
)
  --></li>
  <li ><span class="name"><a href="https://alandix.com/glossary/aibook/bottom-up algorithm">bottom-up algorithm</a></span><!--  Array
(
    [slug] =&gt; bottom-up algorithm
    [parent] =&gt; 
    [name] =&gt; bottom-up algorithm
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; 
)
  --></li>
  <li ><span class="name"><a href="https://alandix.com/glossary/aibook/bottom-up reasoning">bottom-up reasoning</a></span><!--  Array
(
    [slug] =&gt; bottom-up reasoning
    [parent] =&gt; 
    [name] =&gt; bottom-up reasoning
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; 
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/aibook/boundary example">boundary example</a></span><!--  Array
(
    [slug] =&gt; boundary example
    [parent] =&gt; 
    [name] =&gt; boundary example
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; Boundary examples are taken from training data or generated and are at or near the boundary between classes.  It can be useful to identify boundary examples during {{machine learning}} as they are places where mis-classifications are likely and hence may be more heavily weighted to ensure accurate boundary behaviour.   Boundary examples can also be used as part of {{explainable AI}} in order to help a human understand classes created by an AI system.
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/aibook/boundary objects">boundary objects</a></span><!--  Array
(
    [slug] =&gt; boundary objects
    [parent] =&gt; 
    [name] =&gt; boundary objects
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; Boundary objects are physical, digital or conceptual items that can act as a point ofcommunication or collaboration between types of people who otherwsie may have very different goals, knowkedge and backgorunds.  The term was first coined to consider the various groups who work around zoological exhibits in a museum, and examples in that context include the phsyical exhibits and the texonomic structyre used to classify them.   As an example in a digital context, archirects and engineers might collaborate around a CAD drawing even though they understand diffent aspects of it, and use it for different purposes.  In {{synergistic}} AI-human collaboration, we might seek to create bondary objects that have meaning for both human and system.  For example, in {{Query-by-Browsing}} the selected items in the list and SQL query are used differently by AI and human, but are a point of connection between them.
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/aibook/branch and bound search">branch and bound search</a></span><!--  Array
(
    [slug] =&gt; branch and bound search
    [parent] =&gt; 
    [name] =&gt; branch and bound search
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; Branch and bound search is used in {{optimisation}} situations for {{tree seadch}} or {{{graph search}} when there is a cost associated with the path taken to reach a solution, and we want to find the solution with the least path cost.  The path cost is often {{additive}}, for example minimising the number of moves to get to a solution, distances between locations along a geographic route, where each move in a puzzle has a cost.  However, branch and bound search does not rely on this, merely that the costs of a path are {{monotonic}}, that isnthey increase if the path gets longer.  The key insight of branch and bound search is that once we have found best-yet solution, we can {{prune}} whole branches for which the path to the branch node exceeds the current best solution.  In addition, the path cost to a node can be used as a {{heuristic}} to order the {{open list}} and choose which nodes to explore first.
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/aibook/branching factor">branching factor</a></span><!--  Array
(
    [slug] =&gt; branching factor
    [parent] =&gt; 
    [name] =&gt; branching factor
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; In a {{tree} the branching factor of a node refers to the number of branches/children of the node.  For the tree as a whole this may be fixed (e.g. a branching factor of 2 for a {{binary tree}}) or may vary between nodes (as is often the case in {{search trees}}).  One may therefore talk about a maximum or average branching factor of a tree.  For example, for the first move in {{Go}} there are 19x19 possible moves, so the branching factor at the root is 361 (although some are equivalent). &lt;br&gt;&lt;br&gt;. In a {{directed graph}} , that is when the links/arcs between nodes have a direction, one can talk about the forward and backward branching factors.  The former is the number of outward links and the latter the numer of inward links.
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/aibook/breadth first search">breadth first search</a></span><!--  Array
(
    [slug] =&gt; breadth first search
    [parent] =&gt; 
    [name] =&gt; breadth first search
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; In breadth first search of a {{search tree}}, one prgresses by first of all looking at all of the nodes at level 1 (one step form the root), then all the nodes at level 2, etc. 
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/aibook/byzantine conditions">Byzantine conditions</a></span><!--  Array
(
    [slug] =&gt; byzantine conditions
    [parent] =&gt; 
    [name] =&gt; Byzantine conditions
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; Byzantine conditions are apparently rare (but in practice possibly not) confluences of parameters, variables or situations which give rise to unexpectedly bad consequences.  For example, some {{sortling algorithms}} behave particularly badly when given an iniitally fully sorted list of values.  Most algorithms have worst case scenarios, when they behave very badly, but so ling as they are rare this is accpeted.  However, soemtimes the structure of processes that give rise to input data means these rare situations happen surprisingly ooftem -- the case of a sorted list is one.  &lt;br&gt;&lt;br&gt; One way to avoid Byzantine conditions is to insert some form of {{randomness}} or {[stochastic}} choices.  For example, a {[dterministic}} version of {{quicksort}} uses the first element as a separator to divide the rest of the list into those greater and smaller.  This leads precisely to a worst case situation with a sorted list as input -- a Byzantine condition.  If instead the separator is chosen as a {{random}} item from the list, the worst case can still happen, but is vanishingly rare.
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/aibook/c4.5">C4.5</a></span><!--  Array
(
    [slug] =&gt; c4.5
    [parent] =&gt; 
    [name] =&gt; C4.5
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; C4.5 is one of the most well known variants of {{Quinlan}}'s {{ID3}} {{decision tree}} learning algorithm.  C4.5 has a number of variations from basic ID3 including the use of {{normalised information gain}} for non-binary node decisions.
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/aibook/cambridge analytica scandal">Cambridge Analytica scandal</a></span><!--  Array
(
    [slug] =&gt; cambridge analytica scandal
    [parent] =&gt; 
    [name] =&gt; Cambridge Analytica scandal
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; Cambridge Analytica was a counsulting company that infamously used large amounts of data gathered from Facebook to help target political advertising.  The extent of this was uncovered in 2018 including its use in the 2016 Trump presidential campaign.  The use was deemed to have violated {{privacy}} and various {{data protection laws}} and Facebook was subsequently fined in both the UK and USA.  
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/aibook/cartesian coordinates">Cartesian coordinates</a></span><!--  Array
(
    [slug] =&gt; cartesian coordinates
    [parent] =&gt; 
    [name] =&gt; Cartesian coordinates
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; Cartesian coordinates are the rectilinear x,y (2D) or x,y,z (3D) cooridinates we see in graphs and maps.  These are a common way to represent locations on a plane or in space, or pixels in an image, but sometimes alternatives are used, for example, ((radial coordinates}} where (in 2D) each point. is represented as a distance and angle from centre point.  Soemtimes a {{change in representation}} can make mathematical analysis or {{machine learning}} easier.
)
  --></li>
  <li ><span class="name"><a href="https://alandix.com/glossary/aibook/case grammar">case grammar</a></span><!--  Array
(
    [slug] =&gt; case grammar
    [parent] =&gt; 
    [name] =&gt; case grammar
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; 
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/aibook/case-based reasoning">case-based reasoning</a></span><!--  Array
(
    [slug] =&gt; case-based reasoning
    [parent] =&gt; 
    [name] =&gt; case-based reasoning
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; Case-based reasing uses a {{database}} of examples (cases) andwhen it encounters a new situation matces it to the database, retreives the closest past examples, and attempts to understand the current situation in the light of these past cases.  Doing this usually involves identifying the differences between the currrent situation and retreived case and working out the imoacts of this using some other form of {{reasoning}}.
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/aibook/cellular automata">cellular automata</a></span><!--  Array
(
    [slug] =&gt; cellular automata
    [parent] =&gt; 
    [name] =&gt; cellular automata
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; Cellular automata are simple computational elements laid out in a spatial pattern, most commonly a square grid, where each automata's state updates based only on its own state and that of its neighbours.  The most well known example is {{Comway}}'s {{Game of Life}}. Very simple rules can give rise to complex {{emergent behaviour}}.
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/aibook/central example">central example</a></span><!--  Array
(
    [slug] =&gt; central example
    [parent] =&gt; 
    [name] =&gt; central example
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; When. looking at a class or cluster, a central example is one that is far form the edges, so that small {{pertubations}} in the values in the example do not change the classification.  For example, a blackbird is a &lt;em&gt;central example&lt;/em&gt; of a bird, wereas a penguin (lfughtless swims a lot) owuld be a {{boundary example}}.  Cenral exmaples can be useful in {{explainable AI}} when the rules for allocating classes have been generaed using {{machine learning}}.
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/aibook/centroid">centroid</a></span><!--  Array
(
    [slug] =&gt; centroid
    [parent] =&gt; 
    [name] =&gt; centroid
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; The centrid is the cdentre of a group of points as cacuulated using the {{arithmetic mean}} of each of the {{Cartesian coordinates}} of the points.  For example, in 2D sace, if the points have coordinates [ (2,0), (2,2), (3,1), (1,1) ] then the centroid is ( 2, 1 ), because 2 is the mean of the x coordinates: 2, 2, 3 and 1 and 1 is the mean of the y coordinates: 0, 2, 1, 1.
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/aibook/cerebral cortex">cerebral cortex</a></span><!--  Array
(
    [slug] =&gt; cerebral cortex
    [parent] =&gt; 
    [name] =&gt; cerebral cortex
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; The cerebral cortex is the deeply folded outer layer of the brain (also known as {{white matter}}).  It is a large sheet of neurons in six rough layers, with different kinds of neurons in each layer
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/aibook/cern">CERN</a></span><!--  Array
(
    [slug] =&gt; cern
    [parent] =&gt; 
    [name] =&gt; CERN
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; CERN is the international particle accelerator and research centre in Switzerland where the {{World Wide Web}} began.

)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/aibook/certainty factors">certainty factors</a></span><!--  Array
(
    [slug] =&gt; certainty factors
    [parent] =&gt; 
    [name] =&gt; certainty factors
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; Certainty factors are way of performing {{uncertain reasoning}}.  While {{Bayesian reasinong}} demands that belief be precisely coded as a single {{probability}}-like number, confidence factors allow a range with belief for and against a particlar conclusion to build as new evidence is found.   Initially , each initial fact/evidence is initially given a certainty factor where positve in the range +1  to -1, where positive values denote evdience for and negative evidence againts, worth zero meaning no evidence and the extremes +1/-1 meaning incontrovertable evidence for/against).  Derived facts are given two values, a measure of belief and a measure of diselief (each [0,1]) based on evidence, and these are accumulated as further positive or negative evidence arises.   This works well in medical domains where there can be evidence both for and against a diagnosis.    Certainty factors were used in the successful early {{expert system}} {{MYCIN} and still used in {{expert systems}} including {{neural network}} variants.  However, they are less popular now than {{Bayesian methods}} and {{fuzzy reasoning}}.
)
  --></li>
  <li ><span class="name">changes in status</span> - See <span class="has_content"><a href="https://alandix.com/glossary/aibook/status change event">status change event</a></span><!--  Array
(
    [slug] =&gt; changes in status
    [parent] =&gt; 
    [name] =&gt; changes in status
    [canonical] =&gt; 
    [see] =&gt; status change event
    [description] =&gt; 
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/aibook/charniak, eugene">Eugene Charniak</a></span><!--  Array
(
    [slug] =&gt; charniak, eugene
    [parent] =&gt; 
    [name] =&gt; Eugene Charniak
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; Eugene Charniak was a Professor at Brown University and an early AI pioneer.  Aminsgt other thins he worked on the way common sense knowlegde is needed to interpret simple chuldren's stories.
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/aibook/chatbots">chatbots</a></span><!--  Array
(
    [slug] =&gt; chatbots
    [parent] =&gt; 
    [name] =&gt; chatbots
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; Chatbots are a form of software that converses using text-based communication in a (more or less) human-like manner.  The earliest chatbots such as {{ELIZA}} used very simple pattern matching, and more compelx variants of this are widely used for customer support on websites.  Chatbots, such as {{ChatGPT}} powered by {{large language modles} have transformed the fluidity of chatbots.   Speech-based chatbots are also possible, but are more often known by different terms, such as {{conversational agents}}.
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/aibook/chatgpt">ChatGPT</a></span><!--  Array
(
    [slug] =&gt; chatgpt
    [parent] =&gt; 
    [name] =&gt; ChatGPT
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; ChatGPT is a {{chatbot}} based on {{OpenAI}}'s {{large language models}} {{GPT-3}} and {{GPT-4}}.  Early experiences with ChatGPT transformed tpublic and media perc eptions of AI as it is able to produce text that is not only {grammatically correct, but also appears to understand concepts.  That is, it arguably passes the {{Turing test}} as it can be indistinguishable from human text-based chat.  In additon, the ability to create good-enough student essays, news stories and even film-scripts has caused considerable concern about academic integrity, accuracy of news, and job security of creative professionals.
)
  --></li>
  <li ><span class="name">chess</span> - See <span class="has_content"><a href="https://alandix.com/glossary/aibook/computer chess">computer chess</a></span><!--  Array
(
    [slug] =&gt; chess
    [parent] =&gt; 
    [name] =&gt; chess
    [canonical] =&gt; 
    [see] =&gt; computer chess
    [description] =&gt; 
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/aibook/childhood cognitive development">childhood cognitive development</a></span><!--  Array
(
    [slug] =&gt; childhood cognitive development
    [parent] =&gt; 
    [name] =&gt; childhood cognitive development
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; Childhood cognitive development describes the way that small children's ability to think, reason and learn develops from birth to adulthood, in particular the early years.  Many studies of models of {{human reasoning}} focus on adults, but there are often substantial differences of kind not just level in childhood.  Each person develops differently and at a different pace, but general patterns have been studied.  In particular, {{Piaget}}'s work, which was based on studies of his own children and those in an oprphanage, proposed a linear developent of different cignitive abilities, that proceeds at different paces, but is the same for all children.  While {{Piaget}}'s work is still seminal, the details are challenged and there are now more nuanced views of development.
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/aibook/chinese room argument">Chinese room argument</a></span><!--  Array
(
    [slug] =&gt; chinese room argument
    [parent] =&gt; 
    [name] =&gt; Chinese room argument
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; The Chinese room is a {{thought experiment}} created by the philosopher {{John Searle}} to argue that true machine intelligence is impossible,  In partcular it seeks  to counter the {{physical symbol hypothesis}}, and deminstrate that intelligent behaviour is not the same as {{intelligence}.  Searle posits a sealed room wthin which a person who has no understanding of the Chinese languge manipulates tables of symbols using written rules.  The rules are such that whenever a question in Chinese is posted into the room, the person is able to use the rules to create a meaningful response in Chinese to be posted back out.  Those outside the room might believe that someone in the room understands Chienses because the room as whole behaves like this.  However, there is no {{intentionality}}, no true meaning.  The person can perform manipulations that appear to understand Chinese, but there is no actual understanding.  Counters to this are based laregly on the {{system response}}, that the room as a whole can 'understand' even of the person does not.
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/aibook/chunking">chunking</a></span><!--  Array
(
    [slug] =&gt; chunking
    [parent] =&gt; 
    [name] =&gt; chunking
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; Chunking is used in {{cognitive science}} to refer to the way that groups of items can be tretaed as a single unit (a chunk).  This occurs in {{short-term memory}}, where it is a lot eaiser to remember 5 words than 30 random letters.   Chunking is also used to refer to the active process whereby frequent groups of stiumuli, patterns of thought or well practiced movements become treated as a single chunk.  This can be emulated in computational {{cognitive architectures}} including in {{SOAR}}.
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/aibook/clarke, arthur c">Clarke, Arthur C.</a></span><!--  Array
(
    [slug] =&gt; clarke, arthur c
    [parent] =&gt; 
    [name] =&gt; Clarke, Arthur C.
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; Arthur C. Clarke is a science fiction author, who wrote '2001 a Space Odyssey' (novel and film) that includes the {{sentient}} compter {{HAL}}.
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/aibook/clippy">Clippy</a></span><!--  Array
(
    [slug] =&gt; clippy
    [parent] =&gt; 
    [name] =&gt; Clippy
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; Clippy was an {{intelligent agent}} that was deployed for a period in versions of Microsoft Office (officially called Cliipit).  It was depicted using an icon with speach bubbles, that looked like a paper clip with a face.  Clippy would monitor your work and make useful suggestions.  However, the design of the {{user interface}} for Clippy meant it would interupt your work, so very annoying when the suggestion was not needed.  That is it did NOT embody {{appropriate intelligence}}.
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/aibook/closed list">closed list</a></span><!--  Array
(
    [slug] =&gt; closed list
    [parent] =&gt; 
    [name] =&gt; closed list
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; In {{search algorithms}} the closed list records the set nodes that have already been considered.  It is required in {{graph search}} to avoid re-exmaining the same nodes.  This is in contrast to the {{open-list}} which records nodes that still need to be explored.
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/aibook/closed world assumption">closed world assumption</a></span><!--  Array
(
    [slug] =&gt; closed world assumption
    [parent] =&gt; 
    [name] =&gt; closed world assumption
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; When a {{reasoning system}} adopts the closed world assumption, it assumes that all facts (within its domain) are present in its {{knowledge base}}.  That is if a fact is not in the {{knowledge base}}, it is deemed to be false.  The {{logic programming}} language {{Prolog}} adopts a closed world assumption.
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/aibook/closed-loop control">closed-loop control</a></span><!--  Array
(
    [slug] =&gt; closed-loop control
    [parent] =&gt; 
    [name] =&gt; closed-loop control
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; Closed-loop control is a term used in {{cybernetics}} and {{control engineering}} to describe a system that monitors the impact of its outpits and uses this to adjust its behaviour.  A simple exmaple is the Centrifugal governor in steam engines.  As the engine move faster it spins a mechanism with two heavy balls attached. As they spin they rise up opening.a vavlue that reduce the pressure and hence maintains the speed close to a desired level.
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/aibook/cloud computation">cloud computation</a></span><!--  Array
(
    [slug] =&gt; cloud computation
    [parent] =&gt; 
    [name] =&gt; cloud computation
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; Cloud computing is when computation is performed in remote {{data centres}} rather than in known, local computers.  In some cases you have one or more dedicated computers, on some computers (or virtual computers) are started and stopped on request, and sometimes code is completely serverless, executing wherever ther is capacity.  In most cases the physical infrastructure is managed by a separate company who provide {{compute as a service}}.
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/aibook/clustering">clustering</a></span><!--  Array
(
    [slug] =&gt; clustering
    [parent] =&gt; 
    [name] =&gt; clustering
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; Clustering algorithms take data and tries to sort it into clusters each with some form of similar characteristics.  This is like you might do if looking a scatter graph and noticing that the data falls into two or three groups.  Examples of cluster analysis includes {{k-means}} which is based around calculating centroids of each cluster and {{Khonne networks}} (or {{aelf-organsing maps}}) which are a form of neural netwrok where similar data points are classified close to one another a two dimensional grid.  Clustering algorithms are a form of {{unsupevsied learnong}}, but there are also {{semi-supervised}} forms where some or all of the training set has a classification and the clustering algorithm tries to ensure that each cluster has a consistent label, even if there may be several clusters for each label.
)
  --></li>
  <li ><span class="name">co-operating agents</span> - See <span class="has_content"><a href="https://alandix.com/glossary/aibook/cooperating agents">cooperating agents</a></span><!--  Array
(
    [slug] =&gt; co-operating agents
    [parent] =&gt; 
    [name] =&gt; co-operating agents
    [canonical] =&gt; 
    [see] =&gt; cooperating agents
    [description] =&gt; 
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/aibook/cognitive architecture">cognitive architecture</a></span><!--  Array
(
    [slug] =&gt; cognitive architecture
    [parent] =&gt; 
    [name] =&gt; cognitive architecture
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; A  cognitive architecture is a computational model that attempts to emulate real human cognition, but at the level of larger unuts than {[artificial neurla networks}}.  For example, a cognitive architecture may have a single (complex) unit to represent {{working memory}}.  This can be used as a awy to understand {{human cognition}} or as inspiration for artificial human-like intelligence.   Influential example include {{SOAR}} and {{ACT*}} {{ACT-R}}.  The {[regret model}} described in the book also has a form of cognitive architure including units for emotional response and counter-factual reasoning.
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/aibook/coin-weighing problem">coin-weighing problem</a></span><!--  Array
(
    [slug] =&gt; coin-weighing problem
    [parent] =&gt; 
    [name] =&gt; coin-weighing problem
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; Coin-weighing problems and similar puzzles are often used as simple examples for education or as tests for algorithms.  Typical weighing tasks have some number of coins (or other ietms) that should all be identical, but where one (or more) is known to be heavier, lighter or simply diffeernt from the others (in the case of coins usually a counterfeit).  The goal is usually to find the counterfeit(s) using as few weighings as possible.
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/aibook/combinatorial explosion">combinatorial explosion</a></span><!--  Array
(
    [slug] =&gt; combinatorial explosion
    [parent] =&gt; 
    [name] =&gt; combinatorial explosion
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; Combinatorial explosion is the term applied to certain kinds of problem where the complexity grows very rapidly.  This arises when all combinations of, even relatively small, sets of elements need to be considered.  For example, imagine we are looking at choices of flavours in ice-cream sundaes (ignoring how much, it is a very generous gelateria).  If we have two flavours, strawberry and banana, then there are only four options: plain strawberry; plain banana; strawberry and banana; and (for the restrained) none at all.  If there are three flavours, there are 8 options; for four flavours 16; and for a gelateria with a dozen flavours more than 4096.  
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/aibook/common ground">common ground</a></span><!--  Array
(
    [slug] =&gt; common ground
    [parent] =&gt; 
    [name] =&gt; common ground
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; Common ground is a term, coined by Herbert (Herb) Clark and colleagues, used in the theory of cooperative communications.  It refers to the way that two or more people establish a level of shared understanding sufficient for theor purposes as a side effect of communicating to acheive theor goals.  They typically do not try to have complete shared knowledge of everything, nor even spend lots of time at the beginning establishing shared language, but instead build this through unconscious but skillful use of language includuing {{redundancy}} and {{repair}}.  This can be used as a model for human-AI interactions and inter-agent communications.
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/aibook/compas">COMPAS</a></span><!--  Array
(
    [slug] =&gt; compas
    [parent] =&gt; 
    [name] =&gt; COMPAS
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; COMPAS is an AI based system to predict recidivism, that is the likelihood that someone reoffends after leaving prison.  It was traied on US data and has been used, controversially, in US courts as part of parole decsion making.  Analysis has suggested that the outputs may embody a degree of racial {{bias}}.
)
  --></li>
  <li ><span class="name"><a href="https://alandix.com/glossary/aibook/compliant motion">compliant motion</a></span><!--  Array
(
    [slug] =&gt; compliant motion
    [parent] =&gt; 
    [name] =&gt; compliant motion
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; 
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/aibook/computer chess">computer chess</a></span><!--  Array
(
    [slug] =&gt; computer chess
    [parent] =&gt; 
    [name] =&gt; computer chess
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; For many years computer chess was one of the grand challenges in AI, until, in 1996 the {{IBM}} chess computer {{Big Blue}} defeated the chess grandmaster Garry Kasparov.
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/aibook/concept learning">concept learning</a></span><!--  Array
(
    [slug] =&gt; concept learning
    [parent] =&gt; 
    [name] =&gt; concept learning
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; Concept learning is when the coal of learing is to allocate input into one of a number of distinct classifications. Often concept learing is applied to. binary condition 'in category' vs 'not in category', and some algorithms or typically expressed in this binary form.  Examples of concept learning include symbolic algirthms such as {{version spaces}} and  {{ID3}}, and also sub-symbolic technqies including many kinds of {{neural network}} and {{swarm computing}}.
)
  --></li>
  <li ><span class="name"><a href="https://alandix.com/glossary/aibook/confidence">confidence</a></span><!--  Array
(
    [slug] =&gt; confidence
    [parent] =&gt; 
    [name] =&gt; confidence
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; 
)
  --></li>
  <li ><span class="name"><a href="https://alandix.com/glossary/aibook/configuration parameters">configuration parameters</a></span><!--  Array
(
    [slug] =&gt; configuration parameters
    [parent] =&gt; 
    [name] =&gt; configuration parameters
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; 
)
  --></li>
  <li ><span class="name"><a href="https://alandix.com/glossary/aibook/configuration space">configuration space</a></span><!--  Array
(
    [slug] =&gt; configuration space
    [parent] =&gt; 
    [name] =&gt; configuration space
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; 
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/aibook/confusion matrix">confusion matrix</a></span><!--  Array
(
    [slug] =&gt; confusion matrix
    [parent] =&gt; 
    [name] =&gt; confusion matrix
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt;  Confusion matrix is a {{square matrix}} used to assess the accuracy of {{classification}} methods.  The rows are labelled by the class given by the {{classifier}} and the columns are labelled by the actual class.   Each cell is the {{probability}} (or sometimes {{percentage}} or {{freqeuncy}}) with with which an example that is classified as a partcular class is actually of another.  The diagonal represents correct classifications and the off-diagonal elements errors.  The numbers on a row shoudl add up to 1 (for probabilities) or 100% (for percentages).  In order to turn this information into actual accuracy fifgures, one needs to know about the {{base rate}} for each class.  For example, the diagonal entries could be largely near 100% (highly accureate) for all but one case, but that case might be the most common in practice.
)
  --></li>
  <li ><span class="name"><a href="https://alandix.com/glossary/aibook/connectionist model">connectionist model</a></span><!--  Array
(
    [slug] =&gt; connectionist model
    [parent] =&gt; 
    [name] =&gt; connectionist model
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; 
)
  --></li>
  <li ><span class="name"><a href="https://alandix.com/glossary/aibook/consequent-driven reasoning">consequent-driven reasoning</a></span><!--  Array
(
    [slug] =&gt; consequent-driven reasoning
    [parent] =&gt; 
    [name] =&gt; consequent-driven reasoning
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; 
)
  --></li>
  <li ><span class="name"><a href="https://alandix.com/glossary/aibook/conservative warnings">conservative warnings</a></span><!--  Array
(
    [slug] =&gt; conservative warnings
    [parent] =&gt; 
    [name] =&gt; conservative warnings
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; 
)
  --></li>
  <li ><span class="name"><a href="https://alandix.com/glossary/aibook/constraint propagation">constraint propagation</a></span><!--  Array
(
    [slug] =&gt; constraint propagation
    [parent] =&gt; 
    [name] =&gt; constraint propagation
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; 
)
  --></li>
  <li ><span class="name"><a href="https://alandix.com/glossary/aibook/constraint satisfaction">constraint satisfaction</a></span><!--  Array
(
    [slug] =&gt; constraint satisfaction
    [parent] =&gt; 
    [name] =&gt; constraint satisfaction
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; 
)
  --></li>
  <li ><span class="name"><a href="https://alandix.com/glossary/aibook/constraints">constraints</a></span><!--  Array
(
    [slug] =&gt; constraints
    [parent] =&gt; 
    [name] =&gt; constraints
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; 
)
  --></li>
  <li ><span class="name"><a href="https://alandix.com/glossary/aibook/context">context</a></span><!--  Array
(
    [slug] =&gt; context
    [parent] =&gt; 
    [name] =&gt; context
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; 
)
  --></li>
  <li ><span class="name"><a href="https://alandix.com/glossary/aibook/context-aware interaction">context-aware interaction</a></span><!--  Array
(
    [slug] =&gt; context-aware interaction
    [parent] =&gt; 
    [name] =&gt; context-aware interaction
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; 
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/aibook/context-free grammar">context-free grammar</a></span><!--  Array
(
    [slug] =&gt; context-free grammar
    [parent] =&gt; 
    [name] =&gt; context-free grammar
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; A contewxt free grammar is described using hierarchical rules.  They are used extensively for formal notations such as computer programming languages.  They can also be applied to {{natural language processing}}, but have limitations   The hierarchical rules show how a {{non-terminal}} term expands into (or is composed of) some combination of {{terminals}} other {{non-terminals}}.  For example, part of the expansion of a Noun_Phrase ({{non-terminal}}) is:  &lt;br&gt; &amp;nbsp; &amp;nbsp; &amp;nbsp; Noun_Phrase &amp;nbsp; &amp;nbsp; := &amp;nbsp; &amp;nbsp; &quot;the&quot; &amp;nbsp; Qualified_Noun&lt;br&gt; On the right 'the' is a {{terminal}}, a single word and the end point of grammtical expansion, whereas Qualified_Noun is a {{non-terminal}}, which could end up being a single word (e.g. &quot;ball&quot;), but also has expansions including possessives, adjectives, etc. (e.g. &quot;Bella's blue ball&quot;).  Note that the rules apply no matter in what context the rule is used, hence 'context free'.  However, whether a rule is applied in the first place does depend on previosuly applied grammatical rules.  For example one would not attempt to apply the rules for Verb after &quot;the&quot;.  
)
  --></li>
  <li ><span class="name"><a href="https://alandix.com/glossary/aibook/context-sensitive grammar">context-sensitive grammar</a></span><!--  Array
(
    [slug] =&gt; context-sensitive grammar
    [parent] =&gt; 
    [name] =&gt; context-sensitive grammar
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; 
)
  --></li>
  <li ><span class="name"><a href="https://alandix.com/glossary/aibook/continuous distribution">continuous distribution</a></span><!--  Array
(
    [slug] =&gt; continuous distribution
    [parent] =&gt; 
    [name] =&gt; continuous distribution
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; 
)
  --></li>
  <li ><span class="name"><a href="https://alandix.com/glossary/aibook/contour following">contour following</a></span><!--  Array
(
    [slug] =&gt; contour following
    [parent] =&gt; 
    [name] =&gt; contour following
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; 
)
  --></li>
  <li ><span class="name"><a href="https://alandix.com/glossary/aibook/convariance">convariance</a></span><!--  Array
(
    [slug] =&gt; convariance
    [parent] =&gt; 
    [name] =&gt; convariance
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; 
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/aibook/converge">converge</a></span><!--  Array
(
    [slug] =&gt; converge
    [parent] =&gt; 
    [name] =&gt; converge
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; An {{iterative algorithm}} is said to converge if its results gradually vary less and become closer and closer to some final value.  Not that the former does not imply the latter, for example in the sum 1+1/2+1/3+1/4... the partial answers get closer and closer to each, but the sum does not  converge to a final value.  In contrast the sum 1+1/2+1/4+1/8... does converge as the partial sums values get closer and closer (but never reach) to the value 2.
)
  --></li>
  <li ><span class="name"><a href="https://alandix.com/glossary/aibook/convolutional neural network">convolutional neural network</a></span><!--  Array
(
    [slug] =&gt; convolutional neural network
    [parent] =&gt; 
    [name] =&gt; convolutional neural network
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; 
)
  --></li>
  <li ><span class="name"><a href="https://alandix.com/glossary/aibook/convolutions">convolutions</a></span><!--  Array
(
    [slug] =&gt; convolutions
    [parent] =&gt; 
    [name] =&gt; convolutions
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; 
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/aibook/conway, john">Conway, John</a></span><!--  Array
(
    [slug] =&gt; conway, john
    [parent] =&gt; 
    [name] =&gt; Conway, John
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; John Conway was a mathematician, who, amongst other things, created the {{Game of Life}}, a simple {{cellular automata}} with surprisingly rich behaviour.
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/aibook/cooperating agents">cooperating agents</a></span><!--  Array
(
    [slug] =&gt; cooperating agents
    [parent] =&gt; 
    [name] =&gt; cooperating agents
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; Software {{agents}} may operate entirely independently performing discrete tasks.  However, often multiple agants cooperate on a single task, or share useful infornation with one another.  This can be done using some form of coordinating agent, but more often more loosely coupled.  Communications between agents may be direct using {{messages}} or invoking {{methods}}, or indirect through a shared communication means such as a {{blackboard architecture}}, allowing {{emergent behaviour}} beyond that coded into any individual agent.  The agents may execute within a single machine or in {{distributed AI}} over multiple machines, maybe across the whole internet.  &lt;br&gt;&lt;br&gt;. Multiple agents operating in an environment may affect one another indirectly, and so give rise to unexpected {{emergent behaviour}}, but the term &lt;em&gt;cooperating agents&lt;/em&gt; would normally be reserved for situations where there is some form of explicit or planned cooperation.
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/aibook/correlation matrix">correlation matrix</a></span><!--  Array
(
    [slug] =&gt; correlation matrix
    [parent] =&gt; 
    [name] =&gt; correlation matrix
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; A correlation matrix is fored when you have multiple {{features}} on a stream or set of data, for example multiple {{database}} records or {{time series}}.  The {{correlation cooefficient}} is calculated for each pair of features and set in a matrix.  The matrix is symmetric as the  {{correlation cooefficient}} does not depnd on the order of the features, and the diagonal is the correlation between a feature and itself.  &lt;br&gt;&lt;br&gt; The correlation matrix can be used as a means to visualise the rekationships between featyres, and in oarticukar if there are some goruos of features that are all closely related with one anther.  It can also be used as a {{pre-processing}} step before other analysis such as {{clustering}} of features or {{dimensional reduction}}.
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/aibook/cosine similarity">cosine similarity</a></span><!--  Array
(
    [slug] =&gt; cosine similarity
    [parent] =&gt; 
    [name] =&gt; cosine similarity
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; The cosine similarity measures how close two {{vectors}} are to one another using the cosine of the angle between them.  Unlike a {{Euclidean distance}}, the cosine similarity ignores the size of the two vectors, it just focuses on their direction.  In {{document retrieval}] and {{text processing}}, the cosine silarity is applied to the vector representing the frequencies of words in a document, soemtimes weighted by overalll {{corpus frequency}}.
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/aibook/cost-benefit">cost&ndash;benefit</a></span><!--  Array
(
    [slug] =&gt; cost-benefit
    [parent] =&gt; 
    [name] =&gt; cost&amp;ndash;benefit
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; There is often a cost--benefit balance or cost--benefit {{trade-off}} as decsions that lead to benefots may also have associated costs.  We have to chose whether the benefit is worth the cost.  This is easy if both are certain, in the same units and at the same time.  For example, suppose choice A has a benfit of 100 euro and cost of 50 euro and choice B has a benefit of  70 euro and cost of 10 euro; it is clear that choice A is better.  However, the benefit may depend on the wetaher or the throw of a dice, so that weighing up cost/benefot may involve {{risk}}. It gets more cmplicated when the units of value differ for exmaple if the  benefit is be monetary and the cost about health and well being, such as choosing between a well-paid but boring and stressful job versus an fulfilling job that pays very little.    The costs and benefots might also arose at different times, for example, saving in a pension fund.
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/aibook/counterfactual reasoning">counterfactual reasoning</a></span><!--  Array
(
    [slug] =&gt; counterfactual reasoning
    [parent] =&gt; 
    [name] =&gt; counterfactual reasoning
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; Counterfactual reasoning is when you reason about something that you believe to be false, often about different possible ways the world could be or culd have been.  This might be about past events, &quot;if it hadn't rained it would have been a good day&quot;, or about enitirely fiictional sitiuations &quot;if mice could fly ...&quot;.  In classical {{logic}} this leads to contradicitons, unless one is careful to ensure that the true contradictory facts are not part of our reasoning process.  This is used in mathematics in {{proof by contradiction}}.  In day-to-day counterfactual reasoning, we are more able to compartmentalise, so that in a story about &quot;if mice could fly&quot;, we would avoid thinsg that depended on the (true) fact that mice don't fly.   In {{regret}} we use counterfactual reasoning to weight up the potential advanategs or costs if we had chosen alternative actions &quot;if only I hadn't walked under the ladder ...&quot;.
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/aibook/cousin relationship">cousin relationship</a></span><!--  Array
(
    [slug] =&gt; cousin relationship
    [parent] =&gt; 
    [name] =&gt; cousin relationship
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; The meaning of 'cousin' varies between cultures, but within the {{AI book}} we adopt the definition that X is a cousin of Y if X is the offsping of a sibling of the parent of Y.  For example the daughter of the brother of your mother is a cousin.  In {{Prolog}} we might express this as:  &lt;br&gt; &amp;nbsp; &amp;nbsp; &amp;nbsp; cousin{X,Y)  :-  child(X,A), sibling(A,B), child(Y,B). &lt;br&gt;  Note that one has to be careful in translating these definitions into formal logics, as aspects that are 'obvious' the human may not be to a machine.  For example, if (say in {{Prolog}}), we have parent child relationships, we might think that, &quot;X is a sibling of Y&quot; could be expressed  as: &lt;br&gt; &amp;nbsp; &amp;nbsp; &amp;nbsp; sibling(A,B)  :-  child(A,P), child(B,P). &lt;br&gt;  That is, a sibling of the child of your parent. However, this would mean that  you are your own sibling (see code example tbirds.p), and hence when we put this into the cousin relationship, {{Prolog}} would conclude that your sister is a cousin or even that you are a cousin of yourself.   We think &quot;but of course when you say your sibling is a chld of your parent, you don't mean yourself!&quot;, but this has to be explicitly stated in a formal system. 
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/aibook/credit assignment">credit assignment</a></span><!--  Array
(
    [slug] =&gt; credit assignment
    [parent] =&gt; 
    [name] =&gt; credit assignment
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; Credit assignment in AI is about working out which which rule or parameter should be given the credit (or blame)  when something works well or badly. This is important to enable the syetm to learn, by reinforcing good rules.  However, this can be hard when several rules are involved, hence one of the reaosn for {{explanation-based learning}}.  It can be especially difficult in {{reinformcement learning}} as the reward may occur sometime after the crirtial action with other actiosn in between.  For example, if you take a wrong turn early in a journey it may be some time before you realise, but the last few navigation chocies may nave been reasonable &ndash; it is the first choice that should be given the 'credit' ( this case approbation).
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/aibook/critical point">critical point</a></span><!--  Array
(
    [slug] =&gt; critical point
    [parent] =&gt; 
    [name] =&gt; critical point
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; The term critical point is used in roughly its day-to-day meaning in a variety of contexts, usually where there is some qualitative change in behaviour, such as a phase change with temeprature (e.g. solid/liquid) or crystalisation.  When {{nanoparticle}} deposition is used in {{reservoir computing}} there is a critical point as the susbtrate changes from having no conduction paths to total conduction; at that point it has the most interesting and useful {{quantum}} behaviours.
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/aibook/cross-validation">cross-validation</a></span><!--  Array
(
    [slug] =&gt; cross-validation
    [parent] =&gt; 
    [name] =&gt; cross-validation
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; Cross-validation is when you retain some portion of a training set to verify learning on the rest.  It is a way to test the {{generalisation}} of the resulting {{machine learning}} or {{statistical model}}.  Typically models perform better on the data on which they are trained than on fresh data, but can {{overfit}} the data.  When the goal is to evaluate an algorithm {{k-fold cross validation}} may be used where the data is divided into k subsets (folds) and the cross-validation is perfoemd k tiemes, in each pass withholding out one of the folds from the rtainung set and using ot to validate the outcome..
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/aibook/crowdsourcing">crowdsourcing</a></span><!--  Array
(
    [slug] =&gt; crowdsourcing
    [parent] =&gt; 
    [name] =&gt; crowdsourcing
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; Crowdsourcing is when the general public or other large group collaboratively create a large data set.  This may be done as an exlicit goal of the project, for example Wikipedia editors, or BBC Spring Watch, which asks viewers to post their forst sightings of wildlife in the spring.  It can also occur implicitly as part of other activities, for example in the iSpot nature application people post photos or descriptions of wildlife sightings which other people answre; over time this collectively builds a large knowledge base for the users to consult.  Crowdsourcing is related to {{human computation}}, but the connotations of the latter treat people more like cogs in a machine whereas crowdsourcing usually suggest a level of personal interest..
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/aibook/cuban missile crisis">Cuban missile crisis</a></span><!--  Array
(
    [slug] =&gt; cuban missile crisis
    [parent] =&gt; 
    [name] =&gt; Cuban missile crisis
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; The Cuban missile crisis was a stand-off in 1962 between the USA and USSR over the sitting of Russian nuclear missiles in Cuba, which would place them in close range to mainland USA, like the US ones in Europe were to Russia.  The world got close to nucelaer war, an outcome that was worst for everyone, but for each side they could lose if they backed down first.  It can be seen as a real-life version of te {prisoners' dilemma}}, excpet that happily the US and USSR leadership could talk to one another and find a way to resolve the conflict.  For AI illustrates the importance of {{agent communication}}.
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/aibook/current state representation">current state representation</a></span><!--  Array
(
    [slug] =&gt; current state representation
    [parent] =&gt; 
    [name] =&gt; current state representation
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; In {{robotics}} a current state representation records only a single interpretaion or modle of the world that is incrementally updated with each new observation.  This requires less resources that a historical representation, which keeps track of the hstory of observationsn and actions, but isunable to re-assess past observations given new evidence.
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/aibook/cyberattack">cyberattack</a></span><!--  Array
(
    [slug] =&gt; cyberattack
    [parent] =&gt; 
    [name] =&gt; cyberattack
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; A cyberattack is when digital means are used to infiltrate or damage a compter system.  This may involve some form of computer virus, worm or other self-proposgating code.  It may also involve more human attacks, such as {{phishing}}.  Cyberattacks need not compromise a compter sysyem to casuse damage, in particular {{denial of service}} ({{DoS}}) atacks simply flood a machine with a vast  level of network traffic. the   The outcome of a cyber attack might be to bring donw the system, to extact information, or to leave an infiltaretd system avaiulable to itself perform attacks.  For example, {{IOT}} devices typically have poor security and may be targetted, not to daage them per se, but so that they can be part of a {{bot army}} for {{DoS}} attacks on the real targets.
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/aibook/cyberwarfare">cyberwarfare</a></span><!--  Array
(
    [slug] =&gt; cyberwarfare
    [parent] =&gt; 
    [name] =&gt; cyberwarfare
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; Cyberwarfare is when {[cyber weapons}} or {{cyberattacks}} are used as part of allout war or so more limited conflicts.   More broadly AI and digital technology may be used as part of strategic lanning (such as Isreal's 'Gospel' system) or as part of coventional wepaons (such as GPS-guided smart bombs).   There is great concern about fully {{autonomous weapons}} that make their own decsions about what to attack (and who to kill) without a {{human in the loop}}.
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/aibook/cyberweapons">cyberweapons</a></span><!--  Array
(
    [slug] =&gt; cyberweapons
    [parent] =&gt; 
    [name] =&gt; cyberweapons
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; Cyberweapons is the name of a computational algorithm or device  that can be used to inflict damage.  them ipact may be purely in the digital realm, for examplel, disrupting communications, but may also target physical infrastructure.  Probably the first example of the former was {{Stuxnet}}, which, in the late 2000s, targeted Iranian centrifuges used for nuclear processing.  
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/aibook/cyc project">CYC Project</a></span><!--  Array
(
    [slug] =&gt; cyc project
    [parent] =&gt; 
    [name] =&gt; CYC Project
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; CYC is a long running project to capture and reason about {{general knowledge}}. The CYC project was founded in 1984 by {{Doug Lenat}}.  It uses {{symbolic AI}} explicitly representing knowledge and reasoning and it is the largest endeavour of its kind: OpenCYC 4.0, released in 2012, contains around a quarter of a million concepts and over two million facts about them ins a large {{semantic network}}.  Although it is a general knowledge base, it is divided into microtheories about different topic areas, this structure has enabled some of its successful applications in specific sub-areas.   In some ways {{large-language models}} capture elements of general knowledge due to the vast text sources they process, but lack the explicit {{semantics}} and {{explainability}} of CYC.  In his last paper before his death, Lenat and long-term colleague Gary Marcus, wrote about the ptential of {{LLM}}s to learn frm CYC, in a form of {{neurosymbolic}} AI.
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/aibook/cyclotorsion">cyclotorsion</a></span><!--  Array
(
    [slug] =&gt; cyclotorsion
    [parent] =&gt; 
    [name] =&gt; cyclotorsion
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; In a camera, cyclotorsion is the ability to rotate the camera and thus the horizon
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/aibook/daleks">Daleks</a></span><!--  Array
(
    [slug] =&gt; daleks
    [parent] =&gt; 
    [name] =&gt; Daleks
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; Daleks are a {{robot}}-like enemy of {Dr Who}}.  They have mechanical arms and an eye stalk, but are actually not a robot as such, but a form of prosthesis or {{cyborg}} as there is a creature within.
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/aibook/damasio, antonio">Damasio, Antonio</a></span><!--  Array
(
    [slug] =&gt; damasio, antonio
    [parent] =&gt; 
    [name] =&gt; Damasio, Antonio
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; Antonio Damasio is a neuroscientist and philosopher of consciousness.  His books include &quot;Descartes Error&quot;&quot;, which argues, inter alia, that the source of consciousness is knolwedge of one's internal bodily state.  Crucially, he challenges Descrates' separation of body and mind and argues that {{emotion}} and {{rationality}} are intinately linked.  This is important for AI research as it suggests that not only that {{artificial consciousness}}, if possible, depends on {{artificial emotion}}, but actually any true AI requires this too.
)
  --></li>
  <li ><span class="name">Dartmouth Conference</span> - See <span class="has_content"><a href="https://alandix.com/glossary/aibook/dartmouth workshop">Dartmouth Workshop</a></span><!--  Array
(
    [slug] =&gt; dartmouth conference
    [parent] =&gt; 
    [name] =&gt; Dartmouth Conference
    [canonical] =&gt; 
    [see] =&gt; dartmouth workshop
    [description] =&gt; 
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/aibook/dartmouth workshop">Dartmouth Workshop</a></span><!--  Array
(
    [slug] =&gt; dartmouth workshop
    [parent] =&gt; 
    [name] =&gt; Dartmouth Workshop
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; The Dartmouth Workshop was a meeting of about a dozen researchers in 1956 at Dartmouth College.  It was here that the term 'Artificial Intelligence' was coined, and hence is the official starting point for the discipline, although, the roots cna be traced before this, not least amongst the attendees.
)
  --></li>
  <li ><span class="name">data-driven reasoning</span> - See <span class="has_content"><a href="https://alandix.com/glossary/aibook/forward reasoning">forward reasoning</a></span><!--  Array
(
    [slug] =&gt; data-driven reasoning
    [parent] =&gt; 
    [name] =&gt; data-driven reasoning
    [canonical] =&gt; 
    [see] =&gt; forward reasoning
    [description] =&gt; 
)
  --></li>
  <li ><span class="name"><a href="https://alandix.com/glossary/aibook/data cleaning">data cleaning</a></span><!--  Array
(
    [slug] =&gt; data cleaning
    [parent] =&gt; 
    [name] =&gt; data cleaning
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; 
)
  --></li>
  <li ><span class="name"><a href="https://alandix.com/glossary/aibook/data detectors">data detectors</a></span><!--  Array
(
    [slug] =&gt; data detectors
    [parent] =&gt; 
    [name] =&gt; data detectors
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; 
)
  --></li>
  <li ><span class="name"><a href="https://alandix.com/glossary/aibook/data documentation">data documentation</a></span><!--  Array
(
    [slug] =&gt; data documentation
    [parent] =&gt; 
    [name] =&gt; data documentation
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; 
)
  --></li>
  <li ><span class="name"><a href="https://alandix.com/glossary/aibook/data fusion">data fusion</a></span><!--  Array
(
    [slug] =&gt; data fusion
    [parent] =&gt; 
    [name] =&gt; data fusion
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; 
)
  --></li>
  <li ><span class="name"><a href="https://alandix.com/glossary/aibook/data reduction">data reduction</a></span><!--  Array
(
    [slug] =&gt; data reduction
    [parent] =&gt; 
    [name] =&gt; data reduction
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; 
)
  --></li>
  <li ><span class="name"><a href="https://alandix.com/glossary/aibook/data structure">data structure</a></span><!--  Array
(
    [slug] =&gt; data structure
    [parent] =&gt; 
    [name] =&gt; data structure
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; 
)
  --></li>
  <li ><span class="name"><a href="https://alandix.com/glossary/aibook/data wrangling">data wrangling</a></span><!--  Array
(
    [slug] =&gt; data wrangling
    [parent] =&gt; 
    [name] =&gt; data wrangling
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; 
)
  --></li>
  <li ><span class="name"><a href="https://alandix.com/glossary/aibook/database">database</a></span><!--  Array
(
    [slug] =&gt; database
    [parent] =&gt; 
    [name] =&gt; database
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; 
)
  --></li>
  <li ><span class="name"><a href="https://alandix.com/glossary/aibook/de-bias">de-bias</a></span><!--  Array
(
    [slug] =&gt; de-bias
    [parent] =&gt; 
    [name] =&gt; de-bias
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; 
)
  --></li>
  <li ><span class="name">de-trending</span> - See <span class="has_content"><a href="https://alandix.com/glossary/aibook/trend removal">trend removal</a></span><!--  Array
(
    [slug] =&gt; de-trending
    [parent] =&gt; 
    [name] =&gt; de-trending
    [canonical] =&gt; 
    [see] =&gt; trend removal
    [description] =&gt; 
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/aibook/decibel">decibel</a></span><!--  Array
(
    [slug] =&gt; decibel
    [parent] =&gt; 
    [name] =&gt; decibel
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; The decibel scale is used to measure sound levels.   It is a {{logarithmic scale}} where an increase in 10 decibels means the sound is 10 times louder (as measured by sound energy).  This means that a 20 decibel difference represents a 100 times louder, etc.  The {{logarithmc scale}} is used becasue sounds differ greatly n lodness. For example, a whisper is about 30db, conversation about 60db (one thousand times louder) and nearby car engine 90db (one million times loder than the whisper!).
)
  --></li>
  <li ><span class="name"><a href="https://alandix.com/glossary/aibook/decision support system">decision support system</a></span><!--  Array
(
    [slug] =&gt; decision support system
    [parent] =&gt; 
    [name] =&gt; decision support system
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; 
)
  --></li>
  <li ><span class="name"><a href="https://alandix.com/glossary/aibook/decision tree">decision tree</a></span><!--  Array
(
    [slug] =&gt; decision tree
    [parent] =&gt; 
    [name] =&gt; decision tree
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; 
)
  --></li>
  <li ><span class="name">decision trees</span> - See <span ><a href="https://alandix.com/glossary/aibook/decision tree">decision tree</a></span><!--  Array
(
    [slug] =&gt; decision trees
    [parent] =&gt; 
    [name] =&gt; decision trees
    [canonical] =&gt; 
    [see] =&gt; decision tree
    [description] =&gt; 
)
  --></li>
  <li ><span class="name">deduction</span> - See <span class="has_content"><a href="https://alandix.com/glossary/aibook/deductive reasoning">deductive reasoning</a></span><!--  Array
(
    [slug] =&gt; deduction
    [parent] =&gt; 
    [name] =&gt; deduction
    [canonical] =&gt; 
    [see] =&gt; deductive reasoning
    [description] =&gt; 
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/aibook/declarative knowledge">declarative knowledge</a></span><!--  Array
(
    [slug] =&gt; declarative knowledge
    [parent] =&gt; 
    [name] =&gt; declarative knowledge
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; Declarative knowledge is about rememebered facts about the world, in contrast to {{procedural knowledge}}, which tries tom work out things on the fly.  This distinction applies to {{human reasoning}}, for example, one perosn might recall their seven times table (decarative knowledge), wheras another might recall one or two key multiples and then rapdily add or subtract sevens ({{procedural knowledge}}).  It also applies to {{automated reasoning}} .  In particular {{machine learning}} and {{statistical algorithms}} using the web and other {{big data}} techniques allow declarative reasoning  that may appear intelligent without any real understanding.
)
  --></li>
  <li ><span class="name"><a href="https://alandix.com/glossary/aibook/deductive learning">deductive learning</a></span><!--  Array
(
    [slug] =&gt; deductive learning
    [parent] =&gt; 
    [name] =&gt; deductive learning
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; 
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/aibook/deductive reasoning">deductive reasoning</a></span><!--  Array
(
    [slug] =&gt; deductive reasoning
    [parent] =&gt; 
    [name] =&gt; deductive reasoning
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; Deductive reasoning makes set of rules and known facts to deduce further knowledge.  For example, if you know that grass is green and a field is full of grass, you may deduce that a field will be green.  Deductive reasoning is usually limited to what is definitely known, as opposed to other forms of reasoning, such as {{abduction}}, which may reason from {{incomplete knowledege}}. 
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/aibook/deep blue">Deep Blue</a></span><!--  Array
(
    [slug] =&gt; deep blue
    [parent] =&gt; 
    [name] =&gt; Deep Blue
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; Deep Blue is the {{IBM}}  {{chess computer}} that defeated the chess grandmaster Garry Kasparov in {{1996}}.
)
  --></li>
  <li ><span class="name"><a href="https://alandix.com/glossary/aibook/deep fakes">deep fakes</a></span><!--  Array
(
    [slug] =&gt; deep fakes
    [parent] =&gt; 
    [name] =&gt; deep fakes
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; 
)
  --></li>
  <li ><span class="name"><a href="https://alandix.com/glossary/aibook/deep learning">deep learning</a></span><!--  Array
(
    [slug] =&gt; deep learning
    [parent] =&gt; 
    [name] =&gt; deep learning
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; 
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/aibook/deep neural network">deep neural network</a></span><!--  Array
(
    [slug] =&gt; deep neural network
    [parent] =&gt; 
    [name] =&gt; deep neural network
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; A deep neural network is a {{neural network}} with lots of layers.. Tyoically the layers are also large, in the sense of lots of nodes.  Each layer can also have a differnt size and use different types of learning.  For example, it is common for the first layer to be a {{Restricted Boltzmann Machine}} in order to perofrma {{dimensional reduction}}.  TYpically the inner laters are {{underdetermined}} (many equally good arrangements of weightings); this and theor disrance from the output layer means that {{backpropogation}} or similar learning rules need to have very slow learning rates to avoid instabilities.  In addition, more layers and laregr layers usually req=uires more tarng data.  Together these mean that {{deep learning}} (training of deep neural networks) requires. a lot of training data.  This combination of computational cost and data volume is one of the main reasons that the use deep neural networks were not widely adopted for many years.
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/aibook/definite clause grammar">definite clause grammar</a></span><!--  Array
(
    [slug] =&gt; definite clause grammar
    [parent] =&gt; 
    [name] =&gt; definite clause grammar
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; A definite clause grammar is a way of representing grammars as {{definite clauss}} in {{Prolog}} and related languages.  This allows the grammar to express a level of semantics in the form of 'features' and conditions on these, thus adding a level of {[semantcis}} to the grammar.
)
  --></li>
  <li ><span class="name"><a href="https://alandix.com/glossary/aibook/degrees of freedom (data)">degrees of freedom (data)</a></span><!--  Array
(
    [slug] =&gt; degrees of freedom (data)
    [parent] =&gt; 
    [name] =&gt; degrees of freedom (data)
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; 
)
  --></li>
  <li ><span class="name"><a href="https://alandix.com/glossary/aibook/degrees of freedom (robotics)">degrees of freedom (robotics)</a></span><!--  Array
(
    [slug] =&gt; degrees of freedom (robotics)
    [parent] =&gt; 
    [name] =&gt; degrees of freedom (robotics)
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; 
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/aibook/delta">delta</a></span><!--  Array
(
    [slug] =&gt; delta
    [parent] =&gt; 
    [name] =&gt; delta
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; A delta is a small change.  For files it is a description of the differences between the files and is often used in {{version control}} systenms.  In {{incremental algorithms}} it is the small change in parameter values at each step, for example the updates to weights in {{backpropagation}}.
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/aibook/dendral">DENDRAL</a></span><!--  Array
(
    [slug] =&gt; dendral
    [parent] =&gt; 
    [name] =&gt; DENDRAL
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; DENDRAL was an early {{expert system}}, developed initially in the 1960s, that used {{domain specific knowledge}} about chemical structure in order to work alongside chemists to create new scientifc results.  In some ways this can be seen as a precursor to current scientific discovery systems such as {{AlphaFold}}.
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/aibook/denial of service (dos)">denial of service (DoS)</a></span><!--  Array
(
    [slug] =&gt; denial of service (dos)
    [parent] =&gt; 
    [name] =&gt; denial of service (DoS)
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; A denial of service (DoS) s a form of {{cyberattack}} where large numbers of computers all tryn to access the same computer, often a {{web server}}, menaing that eiether it crashes, or is so slow to respond that legitimate traffic cannot access it.  Typically, the computers performing the atack are themselves machies that have been compromised through a computer virus.
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/aibook/dependent feature">dependent feature</a></span><!--  Array
(
    [slug] =&gt; dependent feature
    [parent] =&gt; 
    [name] =&gt; dependent feature
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; A dependent feature is a feature that only makes sense when some other feature has a specific value.  For example, in a {{database}} of motorvehicles 'number fo doors' is not meaningful if the type of vehicle is 'motorcycle'.  Many {macine learning}} algorithms will perform more efficiently if they take into account this form of dependency, for exmaple, in a {{genetic algorithm}} it is not helpful to {{mutate}} features that cannot have any effect.
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/aibook/dependent variable">dependent variable</a></span><!--  Array
(
    [slug] =&gt; dependent variable
    [parent] =&gt; 
    [name] =&gt; dependent variable
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; In {{linear regression}} and other foms of modelling a {{dependent variable}} is a value that you cannot control or you wish to have as the output of the model.  In contrast to an {{independent variable}}, which is something you can modify or one of the model's inputs.  That is the model is effectively of the form: &lt;br&gt; &amp;nbsp; &amp;nbsp; &amp;nbsp; dependent variables &amp;nbsp; = &amp;nbsp;  model( parameters, independent variables ) &lt;br&gt; where one then tries to fit the parameters.
)
  --></li>
  <li ><span class="name"><a href="https://alandix.com/glossary/aibook/depth first search">depth first search</a></span><!--  Array
(
    [slug] =&gt; depth first search
    [parent] =&gt; 
    [name] =&gt; depth first search
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; 
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/aibook/depth of search tree">depth of search tree</a></span><!--  Array
(
    [slug] =&gt; depth of search tree
    [parent] =&gt; 
    [name] =&gt; depth of search tree
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; The depth of a {{search tree}} is the number of links from the root to the deepest leaf node.
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/aibook/derivational analogy">derivational analogy</a></span><!--  Array
(
    [slug] =&gt; derivational analogy
    [parent] =&gt; 
    [name] =&gt; derivational analogy
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; A derivational analogy is a form of {{analogy}} where the process by which an analogous solution was found is used as well as the solution itself.   This allows new solutions that exploit the way the previous analogous solutuon was found, but may vary the steps.  The new solution iself may look nothing like the old one (taking the {{analogy}} into account), but the solution path does.  For example, soemone might reason &quot;getting out of an escape room is a bit like a maths exam question&quot;.  To solve the maths question they would first of all list all the information given, so in the escape room, they list all the materials that are available.  The solutons for the maths question and the escape room are nothing like each other, but the processes are similar.

)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/aibook/descartes, rene">Descartes, R&eacute;ne</a></span><!--  Array
(
    [slug] =&gt; descartes, rene
    [parent] =&gt; 
    [name] =&gt; Descartes, R&amp;eacute;ne
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; R&amp;eacute;ne Descartes (1596&ndash;1650) was a matehmatician and philosopher famous, amngst other things, for his statement &quot;I think therefore I am&quot;.  His work on geometry led to what we now call {{Cartesian coordinates}} which are used universally in {{graphics}}, {{computer vision}} and {{GIS}}.  He believed that there was a sharp divide between {res cogitans}}, the material that forms mind, and {{res extensa}}, the physical body -- a form of {{dualism}}.  This view is near universally rejected by modern philosophers of {{intelligence}} and {{conscousness}}, who instead espouse {{physicalism}} the belief that all thought and human subjectuve experience cam be explained through physical phenomena.
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/aibook/deterministic">deterministic</a></span><!--  Array
(
    [slug] =&gt; deterministic
    [parent] =&gt; 
    [name] =&gt; deterministic
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; Deterministic phenomena are ones which follow a fixed set of physical or logical rules.  The same initial state will always lead to the same outcomes. In contrast, {{stochastic}} phenomen have some degree of {{random}}, {{probabalistic}} or {{nondeterminsitic}} behaviour. 
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/aibook/deterministic ground">deterministic ground</a></span><!--  Array
(
    [slug] =&gt; deterministic ground
    [parent] =&gt; 
    [name] =&gt; deterministic ground
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; In an {{adaptive user interface}}, determinsitic ground refers to the thinsg tht are not adapted by the system, and therefore can be relied upon to have consistent behaviour.  For example, an adaptive website might have a recommended path suggesting the next page to look at (adaptive), but also have a fixed navigation menu (deterministic ground).
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/aibook/deterministic search">deterministic search</a></span><!--  Array
(
    [slug] =&gt; deterministic search
    [parent] =&gt; 
    [name] =&gt; deterministic search
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; A determinsitic search uses a fixed rule to decide the order in which to visit states in a {{search tree} or {{search graph}}, and thus will always search in the same order given the same data.  In contrast a {{stochastic search}} may take different choices based on {{random}} chance.
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/aibook/dialogue component">dialogue component</a></span><!--  Array
(
    [slug] =&gt; dialogue component
    [parent] =&gt; 
    [name] =&gt; dialogue component
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; The dialogue component of an {{expert system}} is the part that interacts with the user.  This may be in the form of a question and answer dialogue, or may be more tabular, form-based or graphical.
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/aibook/dialogue">dialogue</a></span>- has subentries<ul class="subentries"><li class="has_content"><span class="name"><a href="https://alandix.com/glossary/aibook/dialogue!mixed control">mixed control</a></span><!--  Array
(
    [slug] =&gt; dialogue!mixed control
    [parent] =&gt; dialogue
    [name] =&gt; mixed control
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; In an {{expert system}}, mixed control is when both user and computer system can initiate interactions.  For example, the user might ask questions of the {{expert system}}, but the system might then ask for information it needs or proactively {{notify}} the user when it notices and important change.
)
  --></li>
<li class="has_content"><span class="name"><a href="https://alandix.com/glossary/aibook/dialogue!system control">system control</a></span><!--  Array
(
    [slug] =&gt; dialogue!system control
    [parent] =&gt; dialogue
    [name] =&gt; system control
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; In an {{expert system}}, system control is when the computer system drives the interaction.  The user may supply input, but only when asked.
)
  --></li>
<li class="has_content"><span class="name"><a href="https://alandix.com/glossary/aibook/dialogue!user control">user control</a></span><!--  Array
(
    [slug] =&gt; dialogue!user control
    [parent] =&gt; dialogue
    [name] =&gt; user control
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; In an {{expert system}}, user control is when the user drives interactions.  The system can only respond to questions or actions initiated by the {{user}}.
)
  --></li></ul><!--  Array
(
    [slug] =&gt; dialogue
    [parent] =&gt; 
    [name] =&gt; dialogue
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; The term 'dialogue' in human&ndash;computer interaction refers to the style and ordering of human and comoyter imouts and outputs.  This may be of a linguistic form, such as in a {{chatbot}}, but the term also includes menu-based, pointing, and graohical interactions.
    [subentries] =&gt; Array
        (
            [dialogue!mixed control] =&gt; Array
                (
                    [slug] =&gt; dialogue!mixed control
                    [parent] =&gt; dialogue
                    [name] =&gt; mixed control
                    [canonical] =&gt; 
                    [see] =&gt; 
                    [description] =&gt; In an {{expert system}}, mixed control is when both user and computer system can initiate interactions.  For example, the user might ask questions of the {{expert system}}, but the system might then ask for information it needs or proactively {{notify}} the user when it notices and important change.
                )

            [dialogue!system control] =&gt; Array
                (
                    [slug] =&gt; dialogue!system control
                    [parent] =&gt; dialogue
                    [name] =&gt; system control
                    [canonical] =&gt; 
                    [see] =&gt; 
                    [description] =&gt; In an {{expert system}}, system control is when the computer system drives the interaction.  The user may supply input, but only when asked.
                )

            [dialogue!user control] =&gt; Array
                (
                    [slug] =&gt; dialogue!user control
                    [parent] =&gt; dialogue
                    [name] =&gt; user control
                    [canonical] =&gt; 
                    [see] =&gt; 
                    [description] =&gt; In an {{expert system}}, user control is when the user drives interactions.  The system can only respond to questions or actions initiated by the {{user}}.
                )

        )

)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/aibook/difference engine">Difference Engine</a></span><!--  Array
(
    [slug] =&gt; difference engine
    [parent] =&gt; 
    [name] =&gt; Difference Engine
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; The Difference Engine was an early computer designed by {{Charles Babbage}} in the first half of the 19th Century.  It uses the method of 'finite differences' to calcualte polynomials, and is thus a fixed purpose calculator unlike Babbage's later designs for the {{Analytic Engine}}, the worlds first programmable computer, alas never constructed except in steam-punk fiction.
)
  --></li>
  <li ><span class="name"><a href="https://alandix.com/glossary/aibook/differential">differential</a></span><!--  Array
(
    [slug] =&gt; differential
    [parent] =&gt; 
    [name] =&gt; differential
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; 
)
  --></li>
  <li ><span class="name"><a href="https://alandix.com/glossary/aibook/diffusion models">diffusion models</a></span><!--  Array
(
    [slug] =&gt; diffusion models
    [parent] =&gt; 
    [name] =&gt; diffusion models
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; 
)
  --></li>
  <li ><span class="name"><a href="https://alandix.com/glossary/aibook/digital filtering">digital filtering</a></span><!--  Array
(
    [slug] =&gt; digital filtering
    [parent] =&gt; 
    [name] =&gt; digital filtering
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; 
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/aibook/digital signal processing">digital signal processing</a></span><!--  Array
(
    [slug] =&gt; digital signal processing
    [parent] =&gt; 
    [name] =&gt; digital signal processing
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; Digital signal processing (DSP) refers to the computer processing of time-series data such as radio or sound.  It is in contrast to various forms of analogue signal processing, involving circuits with fixed components such as resistors and transistors.  Some forms of DSP effectively emulate the largely linear processing of theor analogue predecessors, band-pass filters that tune a singal to a particular frequency.  However, it is also possible to produce more complex filters, for example those that speed up speech without increasing the pitch.
)
  --></li>
  <li ><span class="name"><a href="https://alandix.com/glossary/aibook/digitization">digitization</a></span><!--  Array
(
    [slug] =&gt; digitization
    [parent] =&gt; 
    [name] =&gt; digitization
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; 
)
  --></li>
  <li ><span class="name"><a href="https://alandix.com/glossary/aibook/dimension reduction">dimension reduction</a></span><!--  Array
(
    [slug] =&gt; dimension reduction
    [parent] =&gt; 
    [name] =&gt; dimension reduction
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; 
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/aibook/directed graph">directed graph</a></span><!--  Array
(
    [slug] =&gt; directed graph
    [parent] =&gt; 
    [name] =&gt; directed graph
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; A directed graph is a network of nodes and connections (often called edges), where each edge has a direction.  This is rather like a city with a one-way system.  For example, a graph with three nodes A, B and C, might have the following dirceted edges: A&amp;rarr;B, B&amp;rarr;A and B&amp;rarr;C.  Note how there are links between A and B in both directions; there is a link going from B to C, but not the other way round; and there is no link at all between A and C.  Directed graphs often arise in {{games}}  where moves take one from one gae state to anther, but often are not symetric; so that you can't always reverse your last move.&lt;br&gt;&lt;br&gt; A directed graph is in contrast to an {{undirected graph}} where edges have no direction and effectively link both ways.
)
  --></li>
  <li ><span class="name"><a href="https://alandix.com/glossary/aibook/disambiguation">disambiguation</a></span><!--  Array
(
    [slug] =&gt; disambiguation
    [parent] =&gt; 
    [name] =&gt; disambiguation
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; 
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/aibook/discrete distributions">discrete distributions</a></span><!--  Array
(
    [slug] =&gt; discrete distributions
    [parent] =&gt; 
    [name] =&gt; discrete distributions
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; A discrete distrbution is a probability distribution where the underlying data are {{discrete}} such as integers or a finite set of values.  Examples include the probabilities of different eye colours in a population ({{categorical}} underlying data), or the score after trou=wing a pair of dice ({{numerical}} underlying data).
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/aibook/distributed ai">distributed AI</a></span><!--  Array
(
    [slug] =&gt; distributed ai
    [parent] =&gt; 
    [name] =&gt; distributed AI
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; Distributed AI is where several agents (robots or software), each with AI capability interact with one another communicating and/or acting in a shared environment.  This may involve collaboratively solving a problem such as {{factory robots}} reconfiguring to deal with changing orders or faults.  Alternatively may be a looser form of interaction such as multiple {{autonomous vehicles}} in a busy urban environment.  Note the term is not usually applied when a single AI algorithm is employs {{parallel processing}}, not for {{ensemble methods}} or {{swarm computing}} employing vast numbers of simple agents.
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/aibook/distributed cognition">distributed cognition</a></span><!--  Array
(
    [slug] =&gt; distributed cognition
    [parent] =&gt; 
    [name] =&gt; distributed cognition
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; Distributed cognition refers to the way that humans (and other animals) effectibely think using aspects of their enviornment inclduing interactions with other people.  Ths may be explciit, such as using a calcuator to help arithmetic.  It may also be implicit such as knowing were one is in the process of assembling a piece of flat-pack furnture becasue you can see the part-assembled piece in front of you.  Aspects of distributed cognition can form the inspiration for {{distributed AI}} and also for understanding the behaviour of {{hybrid human-AI}} systems.
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/aibook/divide and conquer">divide and conquer</a></span><!--  Array
(
    [slug] =&gt; divide and conquer
    [parent] =&gt; 
    [name] =&gt; divide and conquer
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; Divide and conquer is a general {{problem solving}} technique to take a large prolem and braek it into smaller pieces, ecah of which is then easier to solve.  In computer algorithms, the same divide and conquer approach is usually performed {{recursively}} on the parts.  In a {{distributed computing}} setting, such as {{MapReduce}}, different processors may attack different sub-problems.  In an {{agent}}-base dplatform, different agents may divide up a problem beased on their expertise.
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/aibook/document retrieval">document retrieval</a></span><!--  Array
(
    [slug] =&gt; document retrieval
    [parent] =&gt; 
    [name] =&gt; document retrieval
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; Document retrieval (or {{text retreival}} refers to the algorithms and methods used ot query and loacte doucments in large information systems.  In a structured database queries often use indices for specific keys, but within a document retreival system a {{free text index}} is usually required.  More complex {{text processing}} may be used to computer similarities between documents or looking at human usage patterns for {{recommender systems}} to suggest potential documents.
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/aibook/domain-independent knowledge">domain-independent knowledge</a></span><!--  Array
(
    [slug] =&gt; domain-independent knowledge
    [parent] =&gt; 
    [name] =&gt; domain-independent knowledge
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; Domain-independent knowledge is the kind of {{common sense}} or {{general knowledge}} that one has without studying a specific topic or area of expertise.  Simple domain-independent knowledge, such as the idea of generalising firm past examples, may be implicitly built inot the way algoritms work.  More extensive domain-independent knowledge, such as the fact that cats chase birds, is far harder for machines, yet simple for humans.  The long standing {{CYC}} project is one attempt to deal with ths and one can arge the {{large language models}} do capture some level of {{general knowledge}} albiet implicitly in the {{deep neural network}} weights.  Domain-independent knowledge is used in contrast with {{domain-dependent knowledge}}, the specialised knowkedge of a specific area. 
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/aibook/domain-specific knowledge">domain-specific knowledge</a></span><!--  Array
(
    [slug] =&gt; domain-specific knowledge
    [parent] =&gt; 
    [name] =&gt; domain-specific knowledge
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; Domain-dependent knowledge is the specialised knowkedge of a specific topic or area of expertise.     Domain-dependent knowledge is used in contrast with {{domain-independent knowledge}}, which is {{common sense}} or {{general knowledge}}.  For example, a medical {{expert system}} may use a lareg mdeical {{ontology}} or use data on the prevalance of diseases to perform {{Bayesian reasonong}}.  While the rules of reasoning are a simple form of {{domain-independent knowledge}}, the {{ontology}} and detailed daat are domain dependent.
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/aibook/dominoes">dominoes</a></span><!--  Array
(
    [slug] =&gt; dominoes
    [parent] =&gt; 
    [name] =&gt; dominoes
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; Dominoes is the name of the small tiles used in a table top game and of the game itself.  Each domino usually has between zero and six dots at either end, but there are sets with more.  In the AI Book they are used in a simple board game {{placing dominoes}}, which ignores the numbers on the tiles, to illustrate {{game trees}}
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/aibook/dot-com period">dot-com period</a></span><!--  Array
(
    [slug] =&gt; dot-com period
    [parent] =&gt; 
    [name] =&gt; dot-com period
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; The dot-com years refers to the period from approximately late 1998 to early 2000  when there was an explosion of web start-ups and venture funding for commercial hi-tech activity.  It was fuelled by the increasing web user base, but for a period hype overtook the human and technical foundations leading, perhaps inevitably, to a major crash, 'the dot-com bubble'.  The latter was triggerd by the flotation of lastminute.com -- which paradoxically is still a major interbet travel company.  The share price of lastminute rose sharply after listing, but this was soon followed by an even sharper drop.  Confidence in hi-tech companies was hit leading to a near shotdown of funding for hi-tech investments that lasted for several years. 
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/aibook/down-sampling">down-sampling</a></span><!--  Array
(
    [slug] =&gt; down-sampling
    [parent] =&gt; 
    [name] =&gt; down-sampling
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; Down sampling is a term from {{signal processing}} and {{time-series analysis}}.  It refers to reducing the sampling points, often by simply picking every n-th sample.  For example, given the 20 second sampled data: &lt;br&gt; &amp;nbsp; &amp;nbsp; &amp;nbsp; [1.35, 1.37, 1.38, 1.45, 1.42, 1.48, 1.52, 1.51, 1.56], &lt;br&gt;we might take every 3rd item to give one minute samples [1.35, 1.45, 1.52].  More sophisticated down sampling might do some form of {{moving average}} or other {{smoothing}} process before the selection, for example, taking one munute running averages: [1.36, 1.45, 1.53].  &lt;br&gt;&lt;br&gt;Downsamopling is particularly useful as a {{data reduction}} step for slowly changing data.  In a {{window-based algorithm}} it effectively increases how far back the algorithm can look for a given window size.
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/aibook/dreyfus, hubert">Dreyfus, Hubert</a></span><!--  Array
(
    [slug] =&gt; dreyfus, hubert
    [parent] =&gt; 
    [name] =&gt; Dreyfus, Hubert
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; Hubert Dreyfus was a philosopher who was part of the early discussions on the potential for true {{artificial itelligence}}.  He effectvely argued against {{strong AI}} 
 and the {{physical symbol system hypothesis}} suggesting instead that the highly contextual and embodied natiue of intelligence meant that there was a sharp divide between biological human intelligence and computation.
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/aibook/drunkards walk">drunkards walk</a></span><!--  Array
(
    [slug] =&gt; drunkards walk
    [parent] =&gt; 
    [name] =&gt; drunkards walk
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; A drunkards walk refers to a {{stochasric process}} where each step is random, and not based on past history at all.  You can create a simple one-dimensional drunkard's walk by tossing a coin then moving forward a step for each head and back a step for each tail.
)
  --></li>
  <li ><span class="name">DSP </span> - See <span class="has_content"><a href="https://alandix.com/glossary/aibook/digital signal processing">digital signal processing</a></span><!--  Array
(
    [slug] =&gt; dsp
    [parent] =&gt; 
    [name] =&gt; DSP 
    [canonical] =&gt; 
    [see] =&gt; digital signal processing
    [description] =&gt; 
)
  --></li>
  <li ><span class="name"><a href="https://alandix.com/glossary/aibook/dualism">dualism</a></span><!--  Array
(
    [slug] =&gt; dualism
    [parent] =&gt; 
    [name] =&gt; dualism
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; 
)
  --></li>
  <li ><span class="name"><a href="https://alandix.com/glossary/aibook/dynamic function allocation">dynamic function allocation</a></span><!--  Array
(
    [slug] =&gt; dynamic function allocation
    [parent] =&gt; 
    [name] =&gt; dynamic function allocation
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; 
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/aibook/eager">EAGER</a></span><!--  Array
(
    [slug] =&gt; eager
    [parent] =&gt; 
    [name] =&gt; EAGER
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; Eager was an early {{intelligent agent}} that used {{programming-by-demonstration}} to automate user 
)
  --></li>
  <li ><span class="name">EBL</span> - See <span ><a href="https://alandix.com/glossary/aibook/explanation-based learning">explanation-based learning</a></span><!--  Array
(
    [slug] =&gt; ebl
    [parent] =&gt; 
    [name] =&gt; EBL
    [canonical] =&gt; 
    [see] =&gt; explanation-based learning
    [description] =&gt; 
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/aibook/ecg">ECG</a></span><!--  Array
(
    [slug] =&gt; ecg
    [parent] =&gt; 
    [name] =&gt; ECG
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; An ECG (electo-cardiogram) is a device to measure electrical activity in the heart.  The output is the heart beat trace that you will have seen in medical dramas.   Analysing an ECG is challenging for time-series methods as it consist of bursts of actuvity (the heart beat) where the shape of each burst and the spacing between them are each important.  Furthernore the detailed shape of each burst and the length also vary.  Traditional {{time-series analysis}} methods prefer data that is more uniform in structure (even if stochastic), as do some {{neural metods}}.  Some form of multi-level analysis is typically required either explicitly by applying different methods to separate the trace into components and then analyse each, or implicitly, for example in the layers of a {{deep neural network}}.      &lt;br&gt; &lt;a href=&quot;https://commons.wikimedia.org/wiki/File:12-lead_ECG.jpg&quot; target=&quot;_blank&quot;&gt;&lt;img align=&quot;center&quot; src=&quot;https://en.wikipedia.org/wiki/Electrocardiography#/media/File:12-lead_ECG.jpg&quot;&gt;&lt;/a&gt;
)
  --></li>
  <li ><span class="name"><a href="https://alandix.com/glossary/aibook/edge detection">edge detection</a></span><!--  Array
(
    [slug] =&gt; edge detection
    [parent] =&gt; 
    [name] =&gt; edge detection
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; 
)
  --></li>
  <li ><span class="name"><a href="https://alandix.com/glossary/aibook/edge following">edge following</a></span><!--  Array
(
    [slug] =&gt; edge following
    [parent] =&gt; 
    [name] =&gt; edge following
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; 
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/aibook/egocentrism">egocentrism</a></span><!--  Array
(
    [slug] =&gt; egocentrism
    [parent] =&gt; 
    [name] =&gt; egocentrism
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; This refers to the way that young children, and also adults, see thew world from their own perspective and may find it hard even to realise that the world can be seen diffeerntly.  An extreme example of this is the child's game of peek-a-boo, where a young child believes themselves to not be see-able if they can't see.   {{Piaget}} measured this effect by creating a landscape and asking chidlren what could be seen from alocation; young children say that something can be seen if they can see it, whereas older chidlren are able to realise that one side of a hill is not visible from the other.  Although adults are unlikely the think like this at the level of locations and seeing, we still may find it hard to understand that people think differently to us or have fundamentally different viewpoints or values.
)
  --></li>
  <li ><span class="name">eigenvalues</span> - See <span class="has_content"><a href="https://alandix.com/glossary/aibook/eigenvector">eigenvector</a></span><!--  Array
(
    [slug] =&gt; eigenvalues
    [parent] =&gt; 
    [name] =&gt; eigenvalues
    [canonical] =&gt; 
    [see] =&gt; eigenvector
    [description] =&gt; 
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/aibook/eigenvector">eigenvector</a></span><!--  Array
(
    [slug] =&gt; eigenvector
    [parent] =&gt; 
    [name] =&gt; eigenvector
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; An eigenvector of.a {{square matrix}} is a {{vector}} which when multiplied by the matrix ends up as a multiple of itself.  In otherwords, the eigenvector is unchanged except for scale.  Mathentaically, if M is the {{matrix}} and v an eignevector,  Mv = &amp;lambda;v.   The scale factor &amp;lambda; is known as the {{eigenvalue}} of the eigenvector.  Eignevectors have many uses, both directly represented within algorithsm, or as a way to epxlian other processes.  As an an example of the former, if a {{cross-correlation matrix}} is calclated for a dataset, and the eigenvectors with large {{eigenvalues}} computed, these correspond to the directions in which the data has greatest variation, a process called {{principle components analysis}}.  As an example of the altter, Google PageRank agorithm chaxcses link spreading a measure of importance form page to page, however this effectively computes the {{principal eigenvector}} (the one with largest {[eigenvalue}}) of the {{web graph}} represented as a {{connection matrix}}.
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/aibook/eight queens problem">eight queens problem</a></span><!--  Array
(
    [slug] =&gt; eight queens problem
    [parent] =&gt; 
    [name] =&gt; eight queens problem
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; The eight queens problem tries to place eight queens on a standard 8x8 chess board so that no two queens are on the same row, column or diagonal.,
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/aibook/eliza">ELIZA</a></span><!--  Array
(
    [slug] =&gt; eliza
    [parent] =&gt; 
    [name] =&gt; ELIZA
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; ELIZA was an early {{chatbot}} created by{{ Joseph Weizenbaum}}.  It used simple {{template matching}} together with some fixed  prompts which it applied in a semi-random fashion.  The most well known variatnt was based on Rogerian psychology, where the human therapist is trained to relfect back the patient's statements.  For example, if ELIZA detected key terms such as 'mother' in the user's input, it might say &quot;tell me more about your mother&quot;.
)
  --></li>
  <li ><span class="name"><a href="https://alandix.com/glossary/aibook/email filtering agents">email filtering agents</a></span><!--  Array
(
    [slug] =&gt; email filtering agents
    [parent] =&gt; 
    [name] =&gt; email filtering agents
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; 
)
  --></li>
  <li ><span class="name"><a href="https://alandix.com/glossary/aibook/embodied">embodied</a></span><!--  Array
(
    [slug] =&gt; embodied
    [parent] =&gt; 
    [name] =&gt; embodied
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; 
)
  --></li>
  <li ><span class="name"><a href="https://alandix.com/glossary/aibook/emergent behaviour">emergent behaviour</a></span><!--  Array
(
    [slug] =&gt; emergent behaviour
    [parent] =&gt; 
    [name] =&gt; emergent behaviour
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; 
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/aibook/emotion">emotion</a></span><!--  Array
(
    [slug] =&gt; emotion
    [parent] =&gt; 
    [name] =&gt; emotion
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; In day-to-day language, emotion has a broad menaing encomasing all of the ways we assess and feel about the world and peope and thinsg in it:  love, fear, joy, surprise.   In {{psychology}} emotion refers to physiological effects such heart rate, whereas {{feeling}} refers to the interpretation of this in higher-order ways such as 'fear' or 'excitement'.  In addition, {{mood}} refers to the longer term orientation, for exampel, if you are feeling particularly down, or particularly positive a about things.  Crucially in many situations the physiological effect (emotion) preceeds the interpretation ({{feeling}}), and the latter deoends on your {{mood}}.  For example,  if someone burst a balloon your heart will instantly race, but whether this leads to a sense of fun, fear or anger depends on your {{mood} and the situation.   In day-to-day speech both emotion and feeling are often called 'emotion'.  The distinction is important, however, when creating {{intelligent user interfaces}} that seek to interpret or predict human emotional responses.  It can also be useful when considering {{artificial emotion}}, with different ways to emulate the physiological side and the interpretation.
)
  --></li>
  <li ><span class="name"><a href="https://alandix.com/glossary/aibook/empathy">empathy</a></span><!--  Array
(
    [slug] =&gt; empathy
    [parent] =&gt; 
    [name] =&gt; empathy
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; 
)
  --></li>
  <li ><span class="name"><a href="https://alandix.com/glossary/aibook/energy landscape">energy landscape</a></span><!--  Array
(
    [slug] =&gt; energy landscape
    [parent] =&gt; 
    [name] =&gt; energy landscape
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; 
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/aibook/enigma machine">Enigma machine</a></span><!--  Array
(
    [slug] =&gt; enigma machine
    [parent] =&gt; 
    [name] =&gt; Enigma machine
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; The Enigma machine was an encryption device used by the German armed forces during the Second World War.  {{Alan Turing}} and others at Bletchley Park developed the {{Bombe}} a device to help decrypt Enigma machine messages.  This was one of the inspirations for his later work studying computation.
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/aibook/ensemble methods">ensemble methods</a></span>- has subentries<ul class="subentries"><li ><span class="name"><a href="https://alandix.com/glossary/aibook/ensemble methods!heterogeneous">heterogeneous</a></span><!--  Array
(
    [slug] =&gt; ensemble methods!heterogeneous
    [parent] =&gt; ensemble methods
    [name] =&gt; heterogeneous
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; 
)
  --></li>
<li ><span class="name"><a href="https://alandix.com/glossary/aibook/ensemble methods!homogeneous">homogeneous</a></span><!--  Array
(
    [slug] =&gt; ensemble methods!homogeneous
    [parent] =&gt; ensemble methods
    [name] =&gt; homogeneous
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; 
)
  --></li></ul><!--  Array
(
    [slug] =&gt; ensemble methods
    [parent] =&gt; 
    [name] =&gt; ensemble methods
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; Ensemble methods use large numbers of simpler AI units to solve complex problems.  A common example is the {{random forest}} which combines the results of large numbers of {{decsion trees}}, but other kinds of algorithms can be used for the simpler units.  The combination method may be a simple vote, or may involve more complex weighting of the results of individual units.
    [subentries] =&gt; Array
        (
            [ensemble methods!heterogeneous] =&gt; Array
                (
                    [slug] =&gt; ensemble methods!heterogeneous
                    [parent] =&gt; ensemble methods
                    [name] =&gt; heterogeneous
                    [canonical] =&gt; 
                    [see] =&gt; 
                    [description] =&gt; 
                )

            [ensemble methods!homogeneous] =&gt; Array
                (
                    [slug] =&gt; ensemble methods!homogeneous
                    [parent] =&gt; ensemble methods
                    [name] =&gt; homogeneous
                    [canonical] =&gt; 
                    [see] =&gt; 
                    [description] =&gt; 
                )

        )

)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/aibook/entity recognition">entity recognition</a></span><!--  Array
(
    [slug] =&gt; entity recognition
    [parent] =&gt; 
    [name] =&gt; entity recognition
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; Entity recognition refers to the process of finding words, phrases or non-text patterns that represent a thing such as person, number or date.  It typically precedes other forms of {{document processingng}} or {{Intelligent user interfaces}}.  For example, {{data-detectors}} scan emails, web pages or other documents to highlight entities such as {{dates}} and offer shorcuts to things that you can do with them.    Some datatypes such as numbers and date will use some form of {{pattern matching}} such as {[regular expressions}}, others, such as placenames, may involve searching for words or n-grams in public or private databases, a process known as  {{named entity recognition}}.  Often simpler entities, may be used as part of higher-level entity detection; for example an email containing date, location and time could be recognised as refering to an event and thus offer the user the opportunity to add it to their calendar.
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/aibook/entropy">entropy</a></span><!--  Array
(
    [slug] =&gt; entropy
    [parent] =&gt; 
    [name] =&gt; entropy
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; Entropy is a mathematical measure of the {{information}} in a system.  For a single item of data with a number of possible values, each with {{probability}} p&lt;sub&gt;i&lt;/sub&gt;, the entrop is defined as follows:  &lt;br&gt; entropy = &amp;minus; &amp;Sigma; p&lt;sub&gt;i&lt;/sub&gt; &amp;times;  log&lt;sub&gt;2&lt;/sub&gt; p&lt;sub&gt;i&lt;/sub&gt; &lt;br&gt; This is at a maximum when all the p&lt;sub&gt;i&lt;/sub&gt; are equal, which slightly paradoxically can be seen as a state of maximum disorder, in the sense that therre is not a preference to one value or another.  The term originated in the study of thermodyyamincs in physics, but was coined as an information measure by Shannon and Weaver in the late 1940s
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/aibook/epiphenomenon">epiphenomenon</a></span><!--  Array
(
    [slug] =&gt; epiphenomenon
    [parent] =&gt; 
    [name] =&gt; epiphenomenon
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; An epiphenomenon is a secondary effect.  Within the {{philosophy of consciousness}}, it refers to the notion that consciousness does not have a function as such, but has arisen (or its appearance has arisen), as an accident of other processes.
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/aibook/epistemic action">epistemic action</a></span><!--  Array
(
    [slug] =&gt; epistemic action
    [parent] =&gt; 
    [name] =&gt; epistemic action
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; An epistemic action is an action the purpose of which is to gain information.  Low level examples include turning one's head to look at something you glimpses from the corner of your eye,  or openinga. box to see what's inside.  At a higher level you might look something up on the web, or go to a museum.  This is important in {{comouter vision}} as many {{ambiguous images}} resolve quickly if the veiwpoint is changed even slightly by changing camera position, a process known as {{actuve vision}}. 
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/aibook/epistemic interaction">epistemic interaction</a></span><!--  Array
(
    [slug] =&gt; epistemic interaction
    [parent] =&gt; 
    [name] =&gt; epistemic interaction
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; Epistemic interaction refers to the way that {{intelligent user interfaces}} can be designed so that user interactions reveal more information about the users preferences or goals and thus improve future interactions.  For example, if the user hovers over a glossary entry in the text it shows they have some interest in the topic, whereas if the entries are permanently shown in a sidebar it is impossible to know which the user has read.  The former may be prefered as it provides better knowledge, even if marginally less usable.
)
  --></li>
  <li ><span class="name"><a href="https://alandix.com/glossary/aibook/equilibrium">equilibrium</a></span><!--  Array
(
    [slug] =&gt; equilibrium
    [parent] =&gt; 
    [name] =&gt; equilibrium
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; 
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/aibook/euclidean distance">Euclidean distance</a></span><!--  Array
(
    [slug] =&gt; euclidean distance
    [parent] =&gt; 
    [name] =&gt; Euclidean distance
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; Euclidean distance is the straight-line 'as the crows flies' distance between two points.  In two dimensions it can be calculated uisng Pythagoras theorem and the equivalent for higher dimensions.  Alternative distance measures include the {{Manhatten block distance}}, which is the sum of the x and y distances.
)
  --></li>
  <li ><span class="name">European General Data Protection Regulation</span> - See <span class="has_content"><a href="https://alandix.com/glossary/aibook/gdpr">GDPR</a></span><!--  Array
(
    [slug] =&gt; european general data protection regulation
    [parent] =&gt; 
    [name] =&gt; European General Data Protection Regulation
    [canonical] =&gt; 
    [see] =&gt; gdpr
    [description] =&gt; 
)
  --></li>
  <li ><span class="name"><a href="https://alandix.com/glossary/aibook/event">event</a></span><!--  Array
(
    [slug] =&gt; event
    [parent] =&gt; 
    [name] =&gt; event
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; 
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/aibook/evidence-based medicine">evidence-based medicine</a></span><!--  Array
(
    [slug] =&gt; evidence-based medicine
    [parent] =&gt; 
    [name] =&gt; evidence-based medicine
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; Evidence-based medicine is when clinical advice or decsions are based on established empirical studies or other forms of scientific evidence.  At first it seems as though all modern medicine would fall under this category, but there are many areas, for example in psychiatry, where the level of individual difference and complexity of problems makes this kind of evidence hard to gather, or, it may be argued, inappropriate.  Given the range of conditions and symptoms, it is impossible for a doctor or other medical practitioner to keep up to date or remember all the evdience that is published.  This is one of the ways that {{expert systems}} can be an invaluable part of medical practice.
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/aibook/exception sets">exception sets</a></span><!--  Array
(
    [slug] =&gt; exception sets
    [parent] =&gt; 
    [name] =&gt; exception sets
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; When dealing with large data sets, for example, during {{data preparation}} of {{big data}}, one usually attempts to apply generic coded rules to every data item.  However, it can sometimes be hard to craete rules that aply to everything, for example, to deal with &quot;Henry VIII&quot; or &quot;The Artist Formerly Known as Prince&quot; in ruels fpr processing names.  Rather than creating ever more complex, and potentially buggy, rules; one can instead create &lt;em&gt;exception sets&lt;/em&gt; that say, for example, that &quot;Henry VIII&quot; is a valid name, or &quot;treat the age field in record 7932 as 88&quot;.  These exception sets are then applied before generic rules as a form of {{override}}.  This is especially important when the input data set may beuodated and then re-processed.
)
  --></li>
  <li ><span class="name">exclusive or problem</span> - See <span class="has_content"><a href="https://alandix.com/glossary/aibook/xor problem">XOR problem</a></span><!--  Array
(
    [slug] =&gt; exclusive or problem
    [parent] =&gt; 
    [name] =&gt; exclusive or problem
    [canonical] =&gt; 
    [see] =&gt; xor problem
    [description] =&gt; 
)
  --></li>
  <li ><span class="name"><a href="https://alandix.com/glossary/aibook/expert knowledge">expert knowledge</a></span><!--  Array
(
    [slug] =&gt; expert knowledge
    [parent] =&gt; 
    [name] =&gt; expert knowledge
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; 
)
  --></li>
  <li ><span class="name"><a href="https://alandix.com/glossary/aibook/expert system">expert system</a></span>- has subentries<ul class="subentries"><li ><span class="name"><a href="https://alandix.com/glossary/aibook/expert system!brittleness">brittleness</a></span><!--  Array
(
    [slug] =&gt; expert system!brittleness
    [parent] =&gt; expert system
    [name] =&gt; brittleness
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; 
)
  --></li>
<li ><span class="name"><a href="https://alandix.com/glossary/aibook/expert system!hybrid">hybrid</a></span><!--  Array
(
    [slug] =&gt; expert system!hybrid
    [parent] =&gt; expert system
    [name] =&gt; hybrid
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; 
)
  --></li>
<li ><span class="name"><a href="https://alandix.com/glossary/aibook/expert system!limitations">limitations</a></span><!--  Array
(
    [slug] =&gt; expert system!limitations
    [parent] =&gt; expert system
    [name] =&gt; limitations
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; 
)
  --></li>
<li ><span class="name"><a href="https://alandix.com/glossary/aibook/expert system!meta-knowledge">meta-knowledge</a></span><!--  Array
(
    [slug] =&gt; expert system!meta-knowledge
    [parent] =&gt; expert system
    [name] =&gt; meta-knowledge
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; 
)
  --></li>
<li ><span class="name"><a href="https://alandix.com/glossary/aibook/expert system!purpose">purpose</a></span><!--  Array
(
    [slug] =&gt; expert system!purpose
    [parent] =&gt; expert system
    [name] =&gt; purpose
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; 
)
  --></li>
<li ><span class="name"><a href="https://alandix.com/glossary/aibook/expert system!rule tracing">rule tracing</a></span><!--  Array
(
    [slug] =&gt; expert system!rule tracing
    [parent] =&gt; expert system
    [name] =&gt; rule tracing
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; 
)
  --></li>
<li ><span class="name"><a href="https://alandix.com/glossary/aibook/expert system!shell">shell</a></span><!--  Array
(
    [slug] =&gt; expert system!shell
    [parent] =&gt; expert system
    [name] =&gt; shell
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; 
)
  --></li>
<li ><span class="name"><a href="https://alandix.com/glossary/aibook/expert system!verification">verification</a></span><!--  Array
(
    [slug] =&gt; expert system!verification
    [parent] =&gt; expert system
    [name] =&gt; verification
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; 
)
  --></li></ul><!--  Array
(
    [slug] =&gt; expert system
    [parent] =&gt; 
    [name] =&gt; expert system
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; 
    [subentries] =&gt; Array
        (
            [expert system!brittleness] =&gt; Array
                (
                    [slug] =&gt; expert system!brittleness
                    [parent] =&gt; expert system
                    [name] =&gt; brittleness
                    [canonical] =&gt; 
                    [see] =&gt; 
                    [description] =&gt; 
                )

            [expert system!hybrid] =&gt; Array
                (
                    [slug] =&gt; expert system!hybrid
                    [parent] =&gt; expert system
                    [name] =&gt; hybrid
                    [canonical] =&gt; 
                    [see] =&gt; 
                    [description] =&gt; 
                )

            [expert system!limitations] =&gt; Array
                (
                    [slug] =&gt; expert system!limitations
                    [parent] =&gt; expert system
                    [name] =&gt; limitations
                    [canonical] =&gt; 
                    [see] =&gt; 
                    [description] =&gt; 
                )

            [expert system!meta-knowledge] =&gt; Array
                (
                    [slug] =&gt; expert system!meta-knowledge
                    [parent] =&gt; expert system
                    [name] =&gt; meta-knowledge
                    [canonical] =&gt; 
                    [see] =&gt; 
                    [description] =&gt; 
                )

            [expert system!purpose] =&gt; Array
                (
                    [slug] =&gt; expert system!purpose
                    [parent] =&gt; expert system
                    [name] =&gt; purpose
                    [canonical] =&gt; 
                    [see] =&gt; 
                    [description] =&gt; 
                )

            [expert system!rule tracing] =&gt; Array
                (
                    [slug] =&gt; expert system!rule tracing
                    [parent] =&gt; expert system
                    [name] =&gt; rule tracing
                    [canonical] =&gt; 
                    [see] =&gt; 
                    [description] =&gt; 
                )

            [expert system!shell] =&gt; Array
                (
                    [slug] =&gt; expert system!shell
                    [parent] =&gt; expert system
                    [name] =&gt; shell
                    [canonical] =&gt; 
                    [see] =&gt; 
                    [description] =&gt; 
                )

            [expert system!verification] =&gt; Array
                (
                    [slug] =&gt; expert system!verification
                    [parent] =&gt; expert system
                    [name] =&gt; verification
                    [canonical] =&gt; 
                    [see] =&gt; 
                    [description] =&gt; 
                )

        )

)
  --></li>
  <li ><span class="name"><a href="https://alandix.com/glossary/aibook/explainable ai">explainable AI</a></span><!--  Array
(
    [slug] =&gt; explainable ai
    [parent] =&gt; 
    [name] =&gt; explainable AI
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; 
)
  --></li>
  <li ><span class="name"><a href="https://alandix.com/glossary/aibook/explanation component">explanation component</a></span><!--  Array
(
    [slug] =&gt; explanation component
    [parent] =&gt; 
    [name] =&gt; explanation component
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; 
)
  --></li>
  <li ><span class="name"><a href="https://alandix.com/glossary/aibook/explanation-based learning">explanation-based learning</a></span><!--  Array
(
    [slug] =&gt; explanation-based learning
    [parent] =&gt; 
    [name] =&gt; explanation-based learning
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; 
)
  --></li>
  <li ><span class="name"><a href="https://alandix.com/glossary/aibook/explicit confirmation">explicit confirmation</a></span><!--  Array
(
    [slug] =&gt; explicit confirmation
    [parent] =&gt; 
    [name] =&gt; explicit confirmation
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; 
)
  --></li>
  <li ><span class="name"><a href="https://alandix.com/glossary/aibook/exploration-exploitation trade-off">exploration-exploitation trade-off</a></span><!--  Array
(
    [slug] =&gt; exploration-exploitation trade-off
    [parent] =&gt; 
    [name] =&gt; exploration-exploitation trade-off
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; 
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/aibook/exponential decay">exponential decay</a></span><!--  Array
(
    [slug] =&gt; exponential decay
    [parent] =&gt; 
    [name] =&gt; exponential decay
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; Exponential decay is when a numerical value reduces by a fixed proportion in each unit of time (or some other variable).  For example, if the amount of water reduces by a third every day. When this happens the value at time t is K&lt;sub&gt;t&lt;/sup&gt; where K is a constant less than 1, or alternatively e&lt;sup&gt;&ndash;&amp;lambda;t&lt;/sup&gt; where  &amp;lambda; = ln(K), the {{natural logarithm}} of K.  Exponential decay reduces faster than any positive power of 1/t.
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/aibook/exponential growth">exponential growth</a></span><!--  Array
(
    [slug] =&gt; exponential growth
    [parent] =&gt; 
    [name] =&gt; exponential growth
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; Exponential growth is when a numerical value increases by a fixed proportion in each unit of time (or some other variable).  For example, if the value of gold grows by 10% each year.  When this happens the value at time t is K&lt;sub&gt;t&lt;/sup&gt; where K is a constant greater than 1, or alternatively e&lt;sup&gt;&amp;lambda;t&lt;/sup&gt; where  &amp;lambda; = ln(K), the {{natural logarithm}} of K.  Exponential growth increases faster than any positive power of t.
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/aibook/extensional representation">extensional representation</a></span><!--  Array
(
    [slug] =&gt; extensional representation
    [parent] =&gt; 
    [name] =&gt; extensional representation
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; For a {{logical expression}} or {{database query}} the extensional representation is the collection of all values that satisfy the expression or are returned by the query.  This is in contrast to the {{intensional representation}}, which is the formulae of the expression or the query expresson in a notation such as {{SQL}}.
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/aibook/f score">F score</a></span><!--  Array
(
    [slug] =&gt; f score
    [parent] =&gt; 
    [name] =&gt; F score
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; The F score is an {{accuracy}} metric that combines {{precision}} and {{recall}}.  It is defined as the harmonic mean of {{precision}} and {{recall}}, that is: &lt;br&gt;1/F = 0.5*( 1/precision + 1/recall ) &lt;br&gt;. There are variants (F&lt;sub&gt;&amp;beta;&lt;/sub&gt; that weight {{precision}} and {{recall}} differently.  Also not to be confused with the {{F distribution}} used in statistical testing.
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/aibook/facial recognition">facial recognition</a></span><!--  Array
(
    [slug] =&gt; facial recognition
    [parent] =&gt; 
    [name] =&gt; facial recognition
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; Facial recognition refers to {{image processing}} systems thart are able to identify faces in images or video and then identify who the person is, using some form of data bases of faces.  This can be very useful, for exampel, in naming family members on a personal photo album, or preventing a terrorust attack.  However, there are also worries about issues of {{privacy}} and potential misue by authoritarian governments.  These worries have beeb made more severe guven that many facial recognition systems exhibit leels of {{bias}} often being less effective for darker skins and hence more likely to erroneoulsy flag a black person as a suspected criminal.
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/aibook/fast fourier transform">Fast Fourier Transform</a></span><!--  Array
(
    [slug] =&gt; fast fourier transform
    [parent] =&gt; 
    [name] =&gt; Fast Fourier Transform
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; A Fast Fourier Transform (FFT) is, as the name suggests, a fast way to perform a {{Fourier transform}} of {{time-series}} or other sequential data.  It exploits symetries in the cacklulatio of the {{Fourier transform}} in order to reuse intermeidate results.  Whereas a brute force version of {{Fourier transform}} is quandratic in the size of gthe input (N&lt;sup&gt;2&lt;/sup&gt;), the FFT is log-linear (N log&lt;sub&gt;2&lt;/sub&gt;N).  For example, for N=1024, FFT is around 100 times faster (1024/log&lt;sub&gt;2&lt;/sub&gt;(1024)).
)
  --></li>
  <li ><span class="name"><a href="https://alandix.com/glossary/aibook/fault-tolerant">fault-tolerant</a></span><!--  Array
(
    [slug] =&gt; fault-tolerant
    [parent] =&gt; 
    [name] =&gt; fault-tolerant
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; 
)
  --></li>
  <li ><span class="name"><a href="https://alandix.com/glossary/aibook/feasibility">feasibility</a></span><!--  Array
(
    [slug] =&gt; feasibility
    [parent] =&gt; 
    [name] =&gt; feasibility
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; 
)
  --></li>
  <li ><span class="name"><a href="https://alandix.com/glossary/aibook/feasible">feasible</a></span><!--  Array
(
    [slug] =&gt; feasible
    [parent] =&gt; 
    [name] =&gt; feasible
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; 
)
  --></li>
  <li ><span class="name"><a href="https://alandix.com/glossary/aibook/feature interaction">feature interaction</a></span><!--  Array
(
    [slug] =&gt; feature interaction
    [parent] =&gt; 
    [name] =&gt; feature interaction
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; 
)
  --></li>
  <li ><span class="name"><a href="https://alandix.com/glossary/aibook/feature sets">feature sets</a></span><!--  Array
(
    [slug] =&gt; feature sets
    [parent] =&gt; 
    [name] =&gt; feature sets
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; 
)
  --></li>
  <li ><span class="name"><a href="https://alandix.com/glossary/aibook/federated learning">federated learning</a></span><!--  Array
(
    [slug] =&gt; federated learning
    [parent] =&gt; 
    [name] =&gt; federated learning
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; 
)
  --></li>
  <li ><span class="name"><a href="https://alandix.com/glossary/aibook/feedback">feedback</a></span><!--  Array
(
    [slug] =&gt; feedback
    [parent] =&gt; 
    [name] =&gt; feedback
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; 
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/aibook/feeling">feeling</a></span><!--  Array
(
    [slug] =&gt; feeling
    [parent] =&gt; 
    [name] =&gt; feeling
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; In {{psychology}} {{emotion}} refers to physiological effects such heart rate, whereas &lt;em&gt;feeling&lt;/em&gt; refers to the interpretation of this in higher-order ways such as 'fear' or 'excitement'.  Crucially in many situations the physiological effect ({{emotion}}) preceeds the interpretation ({{feeling}}).  For example,  if soemone burst a aballoon your heart will instantly race, butwhether this leads to a sense of fun, fear or anger depends on the situation.   In day-to-day speech both emotion and feeling are often called 'emotion'.  The distinction is important, however, when creating {{intelligent user interfaces}} that seek to interpret or predict human emotional responses.  It can also be useful when considering {{artificial emotion}}, with different ways to emulate the physiological side and the interpretation.
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/aibook/figure-ground separation">figure-ground separation</a></span><!--  Array
(
    [slug] =&gt; figure-ground separation
    [parent] =&gt; 
    [name] =&gt; figure-ground separation
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; In a visual scene or image, figure-ground separation is the process of distinguising the background, for example the room or landscape (&lt;em&gt;ground&lt;/em&gt;) from things in the foreground, such as a person or car (the &lt;em&gt;figure&lt;/em&gt;).  Figure-ground separation is often performed at an early stae in an image-processing pipeline before more complex processing to identify the figure.
)
  --></li>
  <li ><span class="name"><a href="https://alandix.com/glossary/aibook/filter bubbles">filter bubbles</a></span><!--  Array
(
    [slug] =&gt; filter bubbles
    [parent] =&gt; 
    [name] =&gt; filter bubbles
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; 
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/aibook/finite impulse response">finite impulse response</a></span><!--  Array
(
    [slug] =&gt; finite impulse response
    [parent] =&gt; 
    [name] =&gt; finite impulse response
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; In a finite input response (FIR) system observations or events far enough in the past have no effect on the current behaviour of the system.  Typically this is true of systems that use a fixed {{window}} ovef the data of, say, the last N items and base responses only on these.  However, this may also be less obvious.  Some statistical time series analysis techniques ar FIR such as  a {{moving average}} as well as {[window-based}} neural networks.
)
  --></li>
  <li ><span class="name"><a href="https://alandix.com/glossary/aibook/finite state machine">finite state machine</a></span><!--  Array
(
    [slug] =&gt; finite state machine
    [parent] =&gt; 
    [name] =&gt; finite state machine
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; 
)
  --></li>
  <li ><span class="name"><a href="https://alandix.com/glossary/aibook/firewalls">firewalls</a></span><!--  Array
(
    [slug] =&gt; firewalls
    [parent] =&gt; 
    [name] =&gt; firewalls
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; 
)
  --></li>
  <li ><span class="name"><a href="https://alandix.com/glossary/aibook/first-order difference">first-order difference</a></span><!--  Array
(
    [slug] =&gt; first-order difference
    [parent] =&gt; 
    [name] =&gt; first-order difference
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; 
)
  --></li>
  <li ><span class="name"><a href="https://alandix.com/glossary/aibook/first-order predicate calculus">first-order predicate calculus</a></span><!--  Array
(
    [slug] =&gt; first-order predicate calculus
    [parent] =&gt; 
    [name] =&gt; first-order predicate calculus
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; 
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/aibook/fitness function">fitness function</a></span><!--  Array
(
    [slug] =&gt; fitness function
    [parent] =&gt; 
    [name] =&gt; fitness function
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; In {{optimisation}} and forms of {{machine learniong}} the fitness function (also known as objective function or utility function), measures how good a solution or state is.  For example, the fitness function for an image reconstruction may be measured as the sum of the squares of difference between the pixel values of the target and generated images.  Note that the choice of fitness function has important impact on efficiency, for example, if it has hard {{thresholds}} this may make incemental learning hard.  The fitness function can also be a source of {{bias}}, for example, if you choose based on popularity you may effectively embody societal discrimination.
)
  --></li>
  <li ><span class="name"><a href="https://alandix.com/glossary/aibook/fitness landscape">fitness landscape</a></span><!--  Array
(
    [slug] =&gt; fitness landscape
    [parent] =&gt; 
    [name] =&gt; fitness landscape
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; 
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/aibook/fixation point">fixation point</a></span><!--  Array
(
    [slug] =&gt; fixation point
    [parent] =&gt; 
    [name] =&gt; fixation point
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; In a camera, the fixation point is where the camera is 'looking'.  That is the point in the world that is at the centre of the camera image.  
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/aibook/flocking behaviour">flocking behaviour</a></span><!--  Array
(
    [slug] =&gt; flocking behaviour
    [parent] =&gt; 
    [name] =&gt; flocking behaviour
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; Many animals including birds and sheep exhibit flocking behaviour; that is they appear to move as an almost fluid whole.  In Starlings this is called a murmuration, in fish shoaling, and for many insects swarming.  The {{emergent behaiour}} arises because of many local interactions, each bird, animal or insect is aware of its neighbours and seeks to avoid bumping into them whikst alsos taying close; some have an idea ow ehere they woudl like to go or avoid, perhaps becasue they have spotted a good food source or a preditor.  These many small interactions lead to the overall flocking behaviour.  This can be modelled mathematically for biologists to understand the animals and also to be used in {{CGI}} for films.  These behaviours can also be used as an inspiration for computational {{swarm algorithms}}, which use large numbers of indivdiually simple {{computational agents}} to tackle complex problems.
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/aibook/fmri">fMRI</a></span><!--  Array
(
    [slug] =&gt; fmri
    [parent] =&gt; 
    [name] =&gt; fMRI
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; Functional magnetic resonance imaging (fMRI) is a form of brain scanning that uses levels of blood oxygen to determine which parts of the brain are most active at any moment.  This is used as part of medical diagnoses, but also by neuroscientists to map the interconnections between parts of the brain.  This data is in turn being used by AI researchers, incuding the {{Human Brain Project}}, to create artifical models of human brain functioning and to inspire more human-like {{neural network}] algorithms.
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/aibook/focus">focus</a></span><!--  Array
(
    [slug] =&gt; focus
    [parent] =&gt; 
    [name] =&gt; focus
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; In a camera, the focus is the the distance at which objects are most sharp.  It is usually controlled by adjusting the dustance between the lens and the light detector.
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/aibook/forgetful hill climbing">forgetful hill climbing</a></span><!--  Array
(
    [slug] =&gt; forgetful hill climbing
    [parent] =&gt; 
    [name] =&gt; forgetful hill climbing
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; The simplest form of {{hill climbing}} search is 'forgetful', that is at each step one moves to an adjoining or close point  with a better (or best) {{fitness function}}.  The is has the dabger of getting stuck as a {{local maxima}} or reach an impasse if some states are not allowed, for example those thart are not {{feasible}} when there are {[constraints}}.  For this reason, variants of hill climbing may searches keep track of past states (memory) and use {{backtracking}} to reisit them in case of impasse, or to explore alternatives suhc as anext best solution.
)
  --></li>
  <li ><span class="name">forward branching factor</span> - See <span class="has_content"><a href="https://alandix.com/glossary/aibook/branching factor">branching factor</a></span><!--  Array
(
    [slug] =&gt; forward branching factor
    [parent] =&gt; 
    [name] =&gt; forward branching factor
    [canonical] =&gt; 
    [see] =&gt; branching factor
    [description] =&gt; 
)
  --></li>
  <li ><span class="name">forward chaining</span> - See <span class="has_content"><a href="https://alandix.com/glossary/aibook/forward reasoning">forward reasoning</a></span><!--  Array
(
    [slug] =&gt; forward chaining
    [parent] =&gt; 
    [name] =&gt; forward chaining
    [canonical] =&gt; 
    [see] =&gt; forward reasoning
    [description] =&gt; 
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/aibook/forward reasoning">forward reasoning</a></span><!--  Array
(
    [slug] =&gt; forward reasoning
    [parent] =&gt; 
    [name] =&gt; forward reasoning
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; Forward reasoning (also known as data-driven reasoning) is used in {{reasoning systems}} when rules are run 'forwards' in order to work out further things form the knowledge one has.  For example, given &quot;Scrates is a man&quot; and &quot;all men are mortal&quot;, one can work out that &quot;Socrates is a man&quot;, or ina. numerical example, given &quot;x&gt;y&quot;, &quot;y&gt;z&quot;, and &quot;z=3&quot; one can work out in one step that that &quot;x&gt;z&quot; and &quot;y&gt;3&quot; and then in a second reasoning step (through towo paths) that &quot;x&gt;3&quot;.    Forward reasoning is particularly useful when the forwards {{branching factor}} is smaller than the backwards {{branching factor}}, for example when there is a relatively small numbers of known facts and rules to combine, or when there is a need to precomputer derived knwoledge as a pre-computation step in order to speed up reasoning after deployment.
)
  --></li>
  <li ><span class="name"><a href="https://alandix.com/glossary/aibook/foundation models">foundation models</a></span><!--  Array
(
    [slug] =&gt; foundation models
    [parent] =&gt; 
    [name] =&gt; foundation models
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; 
)
  --></li>
  <li ><span class="name"><a href="https://alandix.com/glossary/aibook/four knights problem">four knights problem</a></span><!--  Array
(
    [slug] =&gt; four knights problem
    [parent] =&gt; 
    [name] =&gt; four knights problem
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; 
)
  --></li>
  <li ><span class="name"><a href="https://alandix.com/glossary/aibook/fourier analysis">fourier analysis</a></span><!--  Array
(
    [slug] =&gt; fourier analysis
    [parent] =&gt; 
    [name] =&gt; fourier analysis
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; 
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/aibook/fourier transform">Fourier transform</a></span><!--  Array
(
    [slug] =&gt; fourier transform
    [parent] =&gt; 
    [name] =&gt; Fourier transform
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; A Fourier transform converts {{time series}} or other sequential data into a a combination of sine and cosine waves of different frequencies.  This is similar to the frequency level indicators that are seen in high-end amplifiers, but at a much finer scale.  Typically the signal is split into frequencies using some power of 2, as this can be calculated using a {{Fast Fourier Transform}}, but in principle any range of consecutive frequencies can be used.
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/aibook/fovea">fovea</a></span><!--  Array
(
    [slug] =&gt; fovea
    [parent] =&gt; 
    [name] =&gt; fovea
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; The fovea is the central portion of retina on the back of the eye.  The retina consists of light sensing cells, but they are not uniformly spaced, in fact the majority are in a small area towards the centre of vision, the fovea.  In practice this means that a very small region of your vision, roughly the size of your thumb at arms length, is seen at high resolution with increasingly less resolution as you move away from it.  Your impression of a complete detailed view is constructed by your brain in part as your eye rapidly jumps across the scene (known as {{saccades}}) and in part based on memory and past experience.
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/aibook/fpgas">FPGAs</a></span><!--  Array
(
    [slug] =&gt; fpgas
    [parent] =&gt; 
    [name] =&gt; FPGAs
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; FPGAs (field programable gate arrays) are chips comprised of simple logic functions such as AND/OR/NOT interlinked with 'gates' that can be modified, rather link in a memory chip. By chosing which of these gates are open and which closed, a FPGA can be programmed to perform patterns of logic operations almost aas fast as if they had been hard coded into the logic.  
)
  --></li>
  <li ><span class="name"><a href="https://alandix.com/glossary/aibook/frame problem">frame problem</a></span><!--  Array
(
    [slug] =&gt; frame problem
    [parent] =&gt; 
    [name] =&gt; frame problem
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; 
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/aibook/frankenstein">Frankenstein</a></span><!--  Array
(
    [slug] =&gt; frankenstein
    [parent] =&gt; 
    [name] =&gt; Frankenstein
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; Frankenstein is a book written by {{Mary Shelly}} in 1818.  The name is often used to refer to the Hammer Movies monster, but strictly it is the name of the monster's creator, Victor Frankenstein.  It is story of scientific hubris.  Frankenstein used a combination of body parts animated by electricty to create a new living being.  Frankenstein's initial excitement turns to disgust and fear.  It is often used as a warning as to the dangers of {{robotics}} and AI.  While the monster had no mechanical or electronic parts, so not even a {{cyborg}}, the idea of creating an artificial lifeform has many parallels with {{strong AI}} in particular.
)
  --></li>
  <li ><span class="name"><a href="https://alandix.com/glossary/aibook/frequency domain">frequency domain</a></span><!--  Array
(
    [slug] =&gt; frequency domain
    [parent] =&gt; 
    [name] =&gt; frequency domain
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; 
)
  --></li>
  <li ><span class="name"><a href="https://alandix.com/glossary/aibook/frequency space">frequency space</a></span><!--  Array
(
    [slug] =&gt; frequency space
    [parent] =&gt; 
    [name] =&gt; frequency space
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; 
)
  --></li>
  <li ><span class="name"><a href="https://alandix.com/glossary/aibook/fully connected">fully connected</a></span><!--  Array
(
    [slug] =&gt; fully connected
    [parent] =&gt; 
    [name] =&gt; fully connected
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; 
)
  --></li>
  <li ><span class="name"><a href="https://alandix.com/glossary/aibook/function allocation">function allocation</a></span><!--  Array
(
    [slug] =&gt; function allocation
    [parent] =&gt; 
    [name] =&gt; function allocation
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; 
)
  --></li>
  <li ><span class="name"><a href="https://alandix.com/glossary/aibook/fuzzy logic">fuzzy logic</a></span><!--  Array
(
    [slug] =&gt; fuzzy logic
    [parent] =&gt; 
    [name] =&gt; fuzzy logic
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; 
)
  --></li>
  <li ><span class="name"><a href="https://alandix.com/glossary/aibook/fuzzy reasoning">fuzzy reasoning</a></span><!--  Array
(
    [slug] =&gt; fuzzy reasoning
    [parent] =&gt; 
    [name] =&gt; fuzzy reasoning
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; 
)
  --></li>
  <li ><span class="name"><a href="https://alandix.com/glossary/aibook/fuzzy set theory">fuzzy set theory</a></span><!--  Array
(
    [slug] =&gt; fuzzy set theory
    [parent] =&gt; 
    [name] =&gt; fuzzy set theory
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; 
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/aibook/game of life">Game of Life</a></span><!--  Array
(
    [slug] =&gt; game of life
    [parent] =&gt; 
    [name] =&gt; Game of Life
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; Game of Life is a simple two-state {{cellular automata}} based on a rectangular grid of squares, where each square is colured as active (alive) or a background color (dead or empty).  If an empty cell has three live neighbours it becomes active/alive.  If a live cell has two many live neighbours (4 or more) it is overcrowded and dies.  If it has only one live neighbour it is lonely and also dies.  From these simple rules many dynamic patterns emerge.
)
  --></li>
  <li ><span class="name"><a href="https://alandix.com/glossary/aibook/game playing">game playing</a></span>- has subentries<ul class="subentries"><li ><span class="name"><a href="https://alandix.com/glossary/aibook/game playing!heuristics">heuristics</a></span><!--  Array
(
    [slug] =&gt; game playing!heuristics
    [parent] =&gt; game playing
    [name] =&gt; heuristics
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; 
)
  --></li></ul><!--  Array
(
    [slug] =&gt; game playing
    [parent] =&gt; 
    [name] =&gt; game playing
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; 
    [subentries] =&gt; Array
        (
            [game playing!heuristics] =&gt; Array
                (
                    [slug] =&gt; game playing!heuristics
                    [parent] =&gt; game playing
                    [name] =&gt; heuristics
                    [canonical] =&gt; 
                    [see] =&gt; 
                    [description] =&gt; 
                )

        )

)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/aibook/game theory">game theory</a></span><!--  Array
(
    [slug] =&gt; game theory
    [parent] =&gt; 
    [name] =&gt; game theory
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; Game theory is the mathematical study of games.  A 'game' is defined quite formally and is not about fun or enjoyment, but more about rewards and penalties of game play.
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/aibook/game tree">game tree</a></span><!--  Array
(
    [slug] =&gt; game tree
    [parent] =&gt; 
    [name] =&gt; game tree
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; A game tree is a tree representation of game states and moves.  Nodes in the tree are game states and branches are labelled by the player who made them.  In two-person, turn-taking games the player alternates at each level of the tree.  A particular path through the tree represents a particular playing of the game. 
)
  --></li>
  <li ><span class="name">GAN</span> - See <span ><a href="https://alandix.com/glossary/aibook/generative adversarial network">generative adversarial network</a></span><!--  Array
(
    [slug] =&gt; gan
    [parent] =&gt; 
    [name] =&gt; GAN
    [canonical] =&gt; 
    [see] =&gt; generative adversarial network
    [description] =&gt; 
)
  --></li>
  <li ><span class="name">Gaussian distribution</span> - See <span class="has_content"><a href="https://alandix.com/glossary/aibook/normal distribution">Normal distribution</a></span><!--  Array
(
    [slug] =&gt; gaussian distribution
    [parent] =&gt; 
    [name] =&gt; Gaussian distribution
    [canonical] =&gt; 
    [see] =&gt; normal distribution
    [description] =&gt; 
)
  --></li>
  <li ><span class="name"><a href="https://alandix.com/glossary/aibook/gaussian filter">Gaussian filter</a></span><!--  Array
(
    [slug] =&gt; gaussian filter
    [parent] =&gt; 
    [name] =&gt; Gaussian filter
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; 
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/aibook/gdpr">GDPR</a></span><!--  Array
(
    [slug] =&gt; gdpr
    [parent] =&gt; 
    [name] =&gt; GDPR
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; General Data Protection Regulation (GDPR) is the European Union data protection law 
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/aibook/general knowledge">general knowledge</a></span><!--  Array
(
    [slug] =&gt; general knowledge
    [parent] =&gt; 
    [name] =&gt; general knowledge
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; The technical meaning of 'general knowledge' is just what it says, the kinds of knowledge about the world that we all have: the sky is blue, apples fall down, people generally have two arms and two legs.  This is contrasted with more specialised knowledge about a specifi topic (such as quantum mechanics, or 15th Century Kings of France) or a specific situation (such as the locations of furniture in your home).  Until recently it has been very difficult for AI to deal with problems involving general knowledge, however, {{large language models}} trained on vast qunatities of human text are beginning the change this as are systems such as {{IBM Watson}} whihc use a combination of rule-based approaches, pre-processed data and real time access to large data sources.
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/aibook/generalisation">generalisation</a></span><!--  Array
(
    [slug] =&gt; generalisation
    [parent] =&gt; 
    [name] =&gt; generalisation
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; Generalisation is the way in which systems take specific examples and construct more abstract rules that cover multiple cases.  In simple {{rote learniong}} one takes examples and simply recalls the expected output or repsonse for each example.   When the same example is given the learner returns a correct result, but is unable to deal with any input that wasn't in the {{training set}}.  Techniques for generalisation attempt to seek more abstract patterns, trends or commonalities in the traing to enable them to respond to novel inputs.   The term also has a specific meaning in the {{ACT-R}} {{cognitive architecture}}.
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/aibook/generate and filter">generate and filter</a></span><!--  Array
(
    [slug] =&gt; generate and filter
    [parent] =&gt; 
    [name] =&gt; generate and filter
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; A simple model of {{creativity}} is generate and filter, which can be seen as a human version of {{generate and test}}.  Humans (or AI) create several of potential ideas, more or less good, and then discard those that are less good and pursue further those that are better.  The initial generation process may be entirely random, for example, ink blots, or more guided.  Different kinds of art focus more or less on the initial geberations (e.g. painting) or the critical filterin (e.g. framing a landscape in photography).  Early {{artificial creativity}} often focused on generating good enough options leavung ti to the humanto be the critic and choose the best outputs to display.  {{Generative models}} are improving this generation phase as they are built on vast corpora of past human outputs (of varying quality).  Perhaps the real challenge is for AI to act as the critic, but there are systems that are beginning to tackle this, at east at the level of day-to-day creartivity, such as writing an essay.
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/aibook/generate and test">generate and test</a></span><!--  Array
(
    [slug] =&gt; generate and test
    [parent] =&gt; 
    [name] =&gt; generate and test
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; One way to {{search}} for solution to a problem is simply to randomly or systematically &lt;em&gt;generate&lt;/em&gt; a putative solution and then &lt;em&gt;test&lt;/em&gt; whether it solves the problem.  For example, given a game of Sudoku, one can generate squares with randomdigits in and then check if it satisfies the rules of Sudoku (e.g. each digit occurs exactly once in each row and column).  This is usually a highlyinefficient way to {{search}}, as it ignores all of the structure in the {[search space}}.  However, it can be used when the {{search space}} is small to exhaustively check all possibilities. &lt;br&gt;&lt;br&gt; Generate and test can also be used as a startng point to geberate more compelx algi=orithms, for example, by generating {{partial solutions}}, or using {{heuristics}} to guide the generation process.  It can aslo be used theoretically as a comparison point for other algorithms -- if your new complex algorithm is not performing better than generate and test, you know you need to return to the drawing board! 
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/aibook/generations">generations</a></span><!--  Array
(
    [slug] =&gt; generations
    [parent] =&gt; 
    [name] =&gt; generations
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; In biological populations, different indviduals greo, breed and die ate differnet times, the idea of generation os herev=fore kore an average time from birth to breeding. In a {{genetic algorithm}} the articial process is usually more lock-step, with a population being considered as a group, with {{fitness}} assessment, mutation, breeding and death applied at a single time step.
)
  --></li>
  <li ><span class="name"><a href="https://alandix.com/glossary/aibook/generative">generative</a></span><!--  Array
(
    [slug] =&gt; generative
    [parent] =&gt; 
    [name] =&gt; generative
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; 
)
  --></li>
  <li ><span class="name"><a href="https://alandix.com/glossary/aibook/generative adversarial network">generative adversarial network</a></span><!--  Array
(
    [slug] =&gt; generative adversarial network
    [parent] =&gt; 
    [name] =&gt; generative adversarial network
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; 
)
  --></li>
  <li ><span class="name"><a href="https://alandix.com/glossary/aibook/generative ai">generative AI</a></span><!--  Array
(
    [slug] =&gt; generative ai
    [parent] =&gt; 
    [name] =&gt; generative AI
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; 
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/aibook/genes">genes</a></span><!--  Array
(
    [slug] =&gt; genes
    [parent] =&gt; 
    [name] =&gt; genes
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; Genes are areas of a {{chromosome}} that form a coding unit, raher like a word or sentance in text.  At a low-level genes are related to the production of various proteins, but at a higher level give rise to inherited traits such as hair colour.  In humans, other animals and many plants, genes come in pairs one inherited from each parent, but there are exceptions, for example, those missing on the (rather stunted) male Y chromosome, or occasionally addtional chromosomes.   The pattern of genes in an animal or plant is called the {{genotype}} whereas the way this appaers (say eye color) is called the {{phenotype}}.  Without {DNA}} testing only the {{phenotype}} is observabe and able to make a difference to the ppearance, behaviour and abilities of the creature or plant. &lt;br&gt;&lt;br&gt;.{{Genetic algorithms}} seek to mimic the way genes are inherited from parents and the survival of prodgeny is dependent on derived features ({{phenotype}}). &lt;br&gt;&lt;br&gt; In early genetics, genes were regarded as having isolated and rule-like behaviour: a blue-eye gene and a brown-eye gene ({{genotype}}) giving rise to a brown eyed person ({{phenotype}}).  However, it is now known that the mapoing is far more coplex, with many genes and environmental factors giving rise to the observable  {{phenotype}}, and also complex interactions within the genome: genes turned on and off, and genes that interact stringly often clusterd togetger on the chromosome so thet they are usually inherited as a unit. &lt;br&gt;&lt;br&gt; This more nuanced understanding of biological genes suggests potential insights for {{genetic algorithms}}, although the majority still are based on 19th century genetics.!
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/aibook/genetic algorithm">genetic algorithm</a></span><!--  Array
(
    [slug] =&gt; genetic algorithm
    [parent] =&gt; 
    [name] =&gt; genetic algorithm
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; A genetic algorithm takes inspiration from the proceses of natural selection for biological species.  Indiiduals are defined using a set of paaremeters, effectively artificial {{genes}} -- in the biological analogy, the {{genotype}}.  These are used to assess their {{fitness}} based on whatever criteria are considered optimal and any {{constraints}} -- in the biological analogy, the {{phenotype}}.   The least fit indivdiuals are 'die' (removed from the popukation) and the fitest selected for 'breeding'; typically this process is {{stochastic}}, with fitness being sued to weigh the {{proabability}} of a {{random}} decision as to which die and breed.   Those chosen to reed may be subject to random {{mutation}} and pairs of individuals brought together and theor aetiicial {{genes}} randomly mxed, in a process akin to mating.  In some versions only mutation is used so that the procsses is more like the growth of bacetria or asexual reproduction.  In some varuants the new interbred individuals replace their parents' geeration entirely, in some they are added to it (replacing those that died).
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/aibook/genetic programming">genetic programming</a></span><!--  Array
(
    [slug] =&gt; genetic programming
    [parent] =&gt; 
    [name] =&gt; genetic programming
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; Genetic programming is a form of {{genetic algorithm}} where the states of the individual elemnts in a population are some form of program structure.  This is often represented as  a syntax tree, decision tree, or similar structural form.  {{Mutations}} and {{breeding}} are often specially designed to make well formed programs, and the evaluation function usually involves running the programover test cases.
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/aibook/genotype">genotype</a></span><!--  Array
(
    [slug] =&gt; genotype
    [parent] =&gt; 
    [name] =&gt; genotype
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; In biological {{genetics}}  the genotype is the actual gene structure , in contrast to  the {{pheonotype}}, which is the observable characteristics of the organism.  In computational {{genetic algorithms}} there is a similar difference between the way in which the state is encoded, for example, the locations of appliances in a kitchen, and the way that gives rise to, say, plumbing and wiring issues within the kitchen.  Often the mapping between genotype and {{pheonotype}} is effectively a one-way function, where it is hard to work out the geotype given the phenotype.  However, for a {{genetic algorithm}} to work it is essential that there is at least some pattern to the relationship otherwise it will be no better than a {{Monte Carlo algorithm}}.
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/aibook/gestalt">gestalt</a></span><!--  Array
(
    [slug] =&gt; gestalt
    [parent] =&gt; 
    [name] =&gt; gestalt
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; Getsalt is a term used in the {{psychology}} of {{perception}} to describe the way that overall features emerge as a whole rather than by piecing together parts.  For example you just see a 'face' rather than inferring it is a face based on the presence of a nose, mouth and eyes.  {{Bottom-up}} {{computer vision}} algorithms often do nto operat like this, but certain kinds of {{neural network}} do appear to emulate aspects of human gestalt.
)
  --></li>
  <li ><span class="name"><a href="https://alandix.com/glossary/aibook/gesture recognition">gesture recognition</a></span><!--  Array
(
    [slug] =&gt; gesture recognition
    [parent] =&gt; 
    [name] =&gt; gesture recognition
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; 
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/aibook/gig economy">gig economy</a></span><!--  Array
(
    [slug] =&gt; gig economy
    [parent] =&gt; 
    [name] =&gt; gig economy
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; The gig economy refers to the internet or app-based business models that grew up in the late 2010s incuding taxi-like services such as Uber and delivery services such as Deliveroo.  They offer flexible working for those providing the services, but often poor job security and working conditions.  
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/aibook/gis">GIS</a></span><!--  Array
(
    [slug] =&gt; gis
    [parent] =&gt; 
    [name] =&gt; GIS
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; GIS (geographical information systems) refers to computer systems, applications and methods for storing and processing geograohical information such as maps, landuse, locations of landmarks, and routes.  They typically need special ways of storing information (such as {{quad trees}} ) as one is often interested in retreiving things that are, for example, close to location or contained within a region not just exect matches. 
)
  --></li>
  <li ><span class="name"><a href="https://alandix.com/glossary/aibook/glial cells">glial cells</a></span><!--  Array
(
    [slug] =&gt; glial cells
    [parent] =&gt; 
    [name] =&gt; glial cells
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; 
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/aibook/global maximum">global maximum</a></span><!--  Array
(
    [slug] =&gt; global maximum
    [parent] =&gt; 
    [name] =&gt; global maximum
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; The global maximum is the point in a {[search space}} or {[fitness landsape}} with the highest possible value.   When a high score is good, the global maximum is the {{optimal}} solution.  It is in contrast to a {{local maximum}}, that is a point in the {{search space}} that is larger than any of its neighbours, but not necessarily largest overall.
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/aibook/global minimum">global minimum</a></span><!--  Array
(
    [slug] =&gt; global minimum
    [parent] =&gt; 
    [name] =&gt; global minimum
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; The global minimum is the point in a {[search space}} or {[fitness landsape}} with the lowest possible value.   When a low score is good, the global minimum is the {{optimal}} solution.  It is in contrast to a {{local minimum}}, that is a point in the {{search space}} that is smaller than any of its neighbours, but not necessarily smallest overall.
)
  --></li>
  <li ><span class="name"><a href="https://alandix.com/glossary/aibook/global planning">global planning</a></span><!--  Array
(
    [slug] =&gt; global planning
    [parent] =&gt; 
    [name] =&gt; global planning
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; 
)
  --></li>
  <li ><span class="name"><a href="https://alandix.com/glossary/aibook/global synchronisation of oscillations">global synchronisation of oscillations</a></span><!--  Array
(
    [slug] =&gt; global synchronisation of oscillations
    [parent] =&gt; 
    [name] =&gt; global synchronisation of oscillations
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; 
)
  --></li>
  <li ><span class="name"><a href="https://alandix.com/glossary/aibook/global workspace theory">global workspace theory</a></span><!--  Array
(
    [slug] =&gt; global workspace theory
    [parent] =&gt; 
    [name] =&gt; global workspace theory
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; 
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/aibook/go">Go</a></span><!--  Array
(
    [slug] =&gt; go
    [parent] =&gt; 
    [name] =&gt; Go
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; Go is a board game played with white and black pieces on a 19x19 board.  the rules are simple, but the stregtegy complex.  It was generally regarded as too difficult for AI as the {{branching factor}} is so large.  In 2016, however, {{AlphaGo}} defeated a Go grandmaster using a mixture of traditional AI {{game search}} and {{deep learning}}.
)
  --></li>
  <li ><span class="name"><a href="https://alandix.com/glossary/aibook/goal state">goal state</a></span><!--  Array
(
    [slug] =&gt; goal state
    [parent] =&gt; 
    [name] =&gt; goal state
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; 
)
  --></li>
  <li ><span class="name"><a href="https://alandix.com/glossary/aibook/goal-driven reasoning">goal-driven reasoning</a></span><!--  Array
(
    [slug] =&gt; goal-driven reasoning
    [parent] =&gt; 
    [name] =&gt; goal-driven reasoning
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; 
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/aibook/godel, kurt">G&ouml;del, Kurt</a></span><!--  Array
(
    [slug] =&gt; godel, kurt
    [parent] =&gt; 
    [name] =&gt; G&amp;ouml;del, Kurt
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; G&amp;ouml;del, Kurt is a mathematicain famous for his incompleteness theorem that proved it was impossible to prove everything that is true in mathematics.  There are some tings that are true, but for which there can never be a proof.  That is, there will always be gaps in knowledge that are fundamental, not simply to hard to tackle yet.
)
  --></li>
  <li ><span class="name"><a href="https://alandix.com/glossary/aibook/google">Google</a></span><!--  Array
(
    [slug] =&gt; google
    [parent] =&gt; 
    [name] =&gt; Google
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; 
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/aibook/gpt-3">GPT-3</a></span><!--  Array
(
    [slug] =&gt; gpt-3
    [parent] =&gt; 
    [name] =&gt; GPT-3
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; GPT-3 is a {{large-language model}} developed by {{OpenAI}}, and was influential in powering {{ChatGPT}}, which radically changed expectations as the the capabilities of {{AI}}.  The aparent step-change in capability started a conversaton as to whether any intelligent behaviour could in fact be produced based on sheer volume of data and computation.   GPT-3 can be tuned for different tasks using prompts and is derisgned to be used as a service by other applicatons, this has led to a large growth of applications using GPT-3 as the underlying engine.  However, becasue GPT-3 and its successors are built using lareg volumes of {{web data}} there are concerns for {{privacy}} and {{intellectual property}}, especially as it occasionally could reproduce verbatim portions of input documents.
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/aibook/gpt-4">GPT-4</a></span><!--  Array
(
    [slug] =&gt; gpt-4
    [parent] =&gt; 
    [name] =&gt; GPT-4
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; GPT-4 is the successor of {{GPT-3}}.  It used an order of magnitued more training data an weights in its {{deep neural network}}.  It furthered speculation as to the potential for true {{artifcial general intelligence}}.  Fears about  {{intelectual property}} cotnued, but the abaility to create apparently convincing argments (albeit sometimes including entirely fictitious evidence -- called {{hallucinating}}) led to angst in education as it added to existing problems of essay mills and plagiarism.
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/aibook/gpu">GPU</a></span><!--  Array
(
    [slug] =&gt; gpu
    [parent] =&gt; 
    [name] =&gt; GPU
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; A GPU (graphics processing unit) is a hardware chip designed to optimise graphics operations such as rendering images in gaming.  GPUs apply {{parallel processing}} to multiple streams of data simultaneously and have proved useful in data-procesisng applications beyond graphics, notably in neural networks and other AI algorihms.
)
  --></li>
  <li ><span class="name"><a href="https://alandix.com/glossary/aibook/graceful degradation">graceful degradation</a></span><!--  Array
(
    [slug] =&gt; graceful degradation
    [parent] =&gt; 
    [name] =&gt; graceful degradation
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; 
)
  --></li>
  <li ><span class="name"><a href="https://alandix.com/glossary/aibook/gradient descent">gradient descent</a></span><!--  Array
(
    [slug] =&gt; gradient descent
    [parent] =&gt; 
    [name] =&gt; gradient descent
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; 
)
  --></li>
  <li ><span class="name"><a href="https://alandix.com/glossary/aibook/gradient filters">gradient filters</a></span><!--  Array
(
    [slug] =&gt; gradient filters
    [parent] =&gt; 
    [name] =&gt; gradient filters
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; 
)
  --></li>
  <li ><span class="name"><a href="https://alandix.com/glossary/aibook/gradient operators">gradient operators</a></span><!--  Array
(
    [slug] =&gt; gradient operators
    [parent] =&gt; 
    [name] =&gt; gradient operators
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; 
)
  --></li>
  <li ><span class="name"><a href="https://alandix.com/glossary/aibook/grammar">grammar</a></span>- has subentries<ul class="subentries"><li ><span class="name"><a href="https://alandix.com/glossary/aibook/grammar!case">case</a></span><!--  Array
(
    [slug] =&gt; grammar!case
    [parent] =&gt; grammar
    [name] =&gt; case
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; 
)
  --></li>
<li ><span class="name"><a href="https://alandix.com/glossary/aibook/grammar!context-free">context-free</a></span><!--  Array
(
    [slug] =&gt; grammar!context-free
    [parent] =&gt; grammar
    [name] =&gt; context-free
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; 
)
  --></li>
<li ><span class="name"><a href="https://alandix.com/glossary/aibook/grammar!context-sensitive">context-sensitive</a></span><!--  Array
(
    [slug] =&gt; grammar!context-sensitive
    [parent] =&gt; grammar
    [name] =&gt; context-sensitive
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; 
)
  --></li>
<li ><span class="name"><a href="https://alandix.com/glossary/aibook/grammar!rules">rules</a></span><!--  Array
(
    [slug] =&gt; grammar!rules
    [parent] =&gt; grammar
    [name] =&gt; rules
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; 
)
  --></li>
<li ><span class="name"><a href="https://alandix.com/glossary/aibook/grammar!semantic">semantic</a></span><!--  Array
(
    [slug] =&gt; grammar!semantic
    [parent] =&gt; grammar
    [name] =&gt; semantic
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; 
)
  --></li>
<li ><span class="name"><a href="https://alandix.com/glossary/aibook/grammar!syntactic">syntactic</a></span><!--  Array
(
    [slug] =&gt; grammar!syntactic
    [parent] =&gt; grammar
    [name] =&gt; syntactic
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; 
)
  --></li></ul><!--  Array
(
    [slug] =&gt; grammar
    [parent] =&gt; 
    [name] =&gt; grammar
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; 
    [subentries] =&gt; Array
        (
            [grammar!case] =&gt; Array
                (
                    [slug] =&gt; grammar!case
                    [parent] =&gt; grammar
                    [name] =&gt; case
                    [canonical] =&gt; 
                    [see] =&gt; 
                    [description] =&gt; 
                )

            [grammar!context-free] =&gt; Array
                (
                    [slug] =&gt; grammar!context-free
                    [parent] =&gt; grammar
                    [name] =&gt; context-free
                    [canonical] =&gt; 
                    [see] =&gt; 
                    [description] =&gt; 
                )

            [grammar!context-sensitive] =&gt; Array
                (
                    [slug] =&gt; grammar!context-sensitive
                    [parent] =&gt; grammar
                    [name] =&gt; context-sensitive
                    [canonical] =&gt; 
                    [see] =&gt; 
                    [description] =&gt; 
                )

            [grammar!rules] =&gt; Array
                (
                    [slug] =&gt; grammar!rules
                    [parent] =&gt; grammar
                    [name] =&gt; rules
                    [canonical] =&gt; 
                    [see] =&gt; 
                    [description] =&gt; 
                )

            [grammar!semantic] =&gt; Array
                (
                    [slug] =&gt; grammar!semantic
                    [parent] =&gt; grammar
                    [name] =&gt; semantic
                    [canonical] =&gt; 
                    [see] =&gt; 
                    [description] =&gt; 
                )

            [grammar!syntactic] =&gt; Array
                (
                    [slug] =&gt; grammar!syntactic
                    [parent] =&gt; grammar
                    [name] =&gt; syntactic
                    [canonical] =&gt; 
                    [see] =&gt; 
                    [description] =&gt; 
                )

        )

)
  --></li>
  <li ><span class="name"><a href="https://alandix.com/glossary/aibook/graph identifier">graph identifier</a></span><!--  Array
(
    [slug] =&gt; graph identifier
    [parent] =&gt; 
    [name] =&gt; graph identifier
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; 
)
  --></li>
  <li ><span class="name"><a href="https://alandix.com/glossary/aibook/graph search">graph search</a></span><!--  Array
(
    [slug] =&gt; graph search
    [parent] =&gt; 
    [name] =&gt; graph search
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; 
)
  --></li>
  <li ><span class="name"><a href="https://alandix.com/glossary/aibook/graphical information systems">graphical information systems</a></span><!--  Array
(
    [slug] =&gt; graphical information systems
    [parent] =&gt; 
    [name] =&gt; graphical information systems
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; 
)
  --></li>
  <li ><span class="name"><a href="https://alandix.com/glossary/aibook/graphics processing unit">graphics processing unit</a></span><!--  Array
(
    [slug] =&gt; graphics processing unit
    [parent] =&gt; 
    [name] =&gt; graphics processing unit
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; 
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/aibook/grey matter">grey matter</a></span><!--  Array
(
    [slug] =&gt; grey matter
    [parent] =&gt; 
    [name] =&gt; grey matter
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; Grey matter is the material in the outer parts of of the brain as compared with the {{white matter}} towards the center.  The majority of the cell bodies of neorones is within the grey matter and connections within the grey matter are largely local.  The nerve fibres that connect more distant neorones, extend through the {{white matter}}.
)
  --></li>
  <li ><span class="name"><a href="https://alandix.com/glossary/aibook/grey scale">grey scale</a></span><!--  Array
(
    [slug] =&gt; grey scale
    [parent] =&gt; 
    [name] =&gt; grey scale
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; 
)
  --></li>
  <li ><span class="name"><a href="https://alandix.com/glossary/aibook/grey-scale image">grey-scale image</a></span><!--  Array
(
    [slug] =&gt; grey-scale image
    [parent] =&gt; 
    [name] =&gt; grey-scale image
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; 
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/aibook/grices's conversational maxims">Grices's conversational maxims</a></span><!--  Array
(
    [slug] =&gt; grices's conversational maxims
    [parent] =&gt; 
    [name] =&gt; Grices's conversational maxims
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; As part of his pragmatic theory of language, Paul Grice proposed four maxims that were implicitly adopted during most day-to-day human cooperative interactions: &lt;br&gt;&lt;em&gt;quantity&lt;/em&gt; &amp;ndash; try to give as much information as is necessary, but no more; &lt;br&gt;&lt;em&gt;quality&lt;/em&gt; &amp;ndash; try to offer truthful and reliable information; &lt;br&gt;&lt;em&gt;relation&lt;/em&gt; &amp;ndash; try to offer relevant and pertinent information; &lt;br&gt;&lt;em&gt;manner&lt;/em&gt; &amp;ndash; try to be as clear and broef as possible whilst avoiding ambiguity. &lt;br&gt; The overacrching pronciple is that people try to offer infornation that helps to further the conversation.
)
  --></li>
  <li ><span class="name"><a href="https://alandix.com/glossary/aibook/ground truth">ground truth</a></span><!--  Array
(
    [slug] =&gt; ground truth
    [parent] =&gt; 
    [name] =&gt; ground truth
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; 
)
  --></li>
  <li ><span class="name">Hadoop</span> - See <span class="has_content"><a href="https://alandix.com/glossary/aibook/apache hadoop">Apache Hadoop</a></span><!--  Array
(
    [slug] =&gt; hadoop
    [parent] =&gt; 
    [name] =&gt; Hadoop
    [canonical] =&gt; 
    [see] =&gt; apache hadoop
    [description] =&gt; 
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/aibook/hal 9000">HAL 9000</a></span><!--  Array
(
    [slug] =&gt; hal 9000
    [parent] =&gt; 
    [name] =&gt; HAL 9000
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; HAL 9000 was the computer in the film &quot;2001 A Space Odyssey&quot;, which developed {{emotions}} and {{consciousness}}.  In the film it becomes afraid when it makes a mistake, and then murderously paranoid fearing.  HAL is evetually turned off, but as this happens pleads for its continuing existence.
)
  --></li>
  <li ><span class="name"><a href="https://alandix.com/glossary/aibook/handwriting recognition">handwriting recognition</a></span><!--  Array
(
    [slug] =&gt; handwriting recognition
    [parent] =&gt; 
    [name] =&gt; handwriting recognition
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; 
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/aibook/hard problem of consciousness">hard problem of consciousness</a></span><!--  Array
(
    [slug] =&gt; hard problem of consciousness
    [parent] =&gt; 
    [name] =&gt; hard problem of consciousness
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; The hard problem of consciousness is a term coined by philosopher David Chalmers.  It refers to the bare experience of bieng as opposed to the 'easy' problems of consciousness such as keeping track of one's mental states.
)
  --></li>
  <li ><span class="name"><a href="https://alandix.com/glossary/aibook/hard threshold">hard threshold</a></span><!--  Array
(
    [slug] =&gt; hard threshold
    [parent] =&gt; 
    [name] =&gt; hard threshold
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; 
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/aibook/hardy, thomas">Hardy, Thomas</a></span><!--  Array
(
    [slug] =&gt; hardy, thomas
    [parent] =&gt; 
    [name] =&gt; Hardy, Thomas
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; Thomas Hardy (1840-1928) was a novelists famous for his bleak plotlines.
)
  --></li>
  <li ><span class="name">Herbert A. Simon</span> - See Simon, Herbert<!--  Array
(
    [slug] =&gt; Herbert A. Simon
    [parent] =&gt; 
    [name] =&gt; Herbert A. Simon
    [canonical] =&gt; 
    [see] =&gt; Simon, Herbert
    [description] =&gt; 
)
  --></li>
  <li ><span class="name"><a href="https://alandix.com/glossary/aibook/heteroassociative memory">heteroassociative memory</a></span><!--  Array
(
    [slug] =&gt; heteroassociative memory
    [parent] =&gt; 
    [name] =&gt; heteroassociative memory
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; 
)
  --></li>
  <li ><span class="name"><a href="https://alandix.com/glossary/aibook/heterogeneous events">heterogeneous events</a></span><!--  Array
(
    [slug] =&gt; heterogeneous events
    [parent] =&gt; 
    [name] =&gt; heterogeneous events
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; 
)
  --></li>
  <li ><span class="name"><a href="https://alandix.com/glossary/aibook/heterogeneous sources">heterogeneous sources</a></span><!--  Array
(
    [slug] =&gt; heterogeneous sources
    [parent] =&gt; 
    [name] =&gt; heterogeneous sources
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; 
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/aibook/heuristic evaluation function">heuristic evaluation function</a></span><!--  Array
(
    [slug] =&gt; heuristic evaluation function
    [parent] =&gt; 
    [name] =&gt; heuristic evaluation function
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; An heuristic evaluation function is some form of guess at an intermediate state about the eventual outcome of a search.  This is  used in a {{heuristic search}} to act as a guide as to the best order to explore many different options.  For example, whilst searching for shortest routes between places, we might use the {{Euclidean distance}} between two cities as a heuristic for the eventual shortest road distance well before we explore the potential routes between them.
)
  --></li>
  <li ><span class="name"><a href="https://alandix.com/glossary/aibook/heuristic search">heuristic search</a></span><!--  Array
(
    [slug] =&gt; heuristic search
    [parent] =&gt; 
    [name] =&gt; heuristic search
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; 
)
  --></li>
  <li ><span class="name"><a href="https://alandix.com/glossary/aibook/heuristics">heuristics</a></span><!--  Array
(
    [slug] =&gt; heuristics
    [parent] =&gt; 
    [name] =&gt; heuristics
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; 
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/aibook/hick's law">Hick's law</a></span><!--  Array
(
    [slug] =&gt; hick's law
    [parent] =&gt; 
    [name] =&gt; Hick's law
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; Hick's Law is one of the few results in cognitve psychology with a precise formulation, namely that the time taken to mae a choice increases with the logarithmof the number of choices.  That is, if someone has to make a choice between N alternatves, the reaction time  T is given by:&lt;br&gt; &amp;nbsp; &amp;nbsp; &amp;nbsp; T &amp;nbsp; &amp;prop; &amp;nbsp; log( N ) &lt;br&gt;&lt;br&gt; In fact, the original experiments that gave to this were based on highly structured inputs that enabled 'binarychop' style decison strategies.  Given a long list of randomly sorted alternatives there is at least a linear time to read the list.  However, whether the exact time is logarithmic or linear, the qulaitiave result still holds: the more alterbatuves the longer ot takes a person to kae the decsions. &lt;br&gt;&lt;br&gt; This isimportant when designing intelligent {{human-in-the-loop}} systems, as it is sensible to present muliple alternatives to the human, but not too many.
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/aibook/hidden markov model">hidden Markov model</a></span><!--  Array
(
    [slug] =&gt; hidden markov model
    [parent] =&gt; 
    [name] =&gt; hidden Markov model
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; A hidden Markov model (HMM) extends the {{Markov model}} by adding a hidden state.  This can be used to model natural phenomena, such as weather systems that ave some sort of longer term meteorological state as well as the immediately measurable rain or shine.   Formally, given a {[time series}} or sequence of observable events/tokens, while a simple Markov model has a matrix of {{probabilities}} M&lt;sub&gt;jk&lt;/sub&gt; giving the probability that event j comes after event k; instead the hidde Markov model has more complex matrix H&lt;sub&gt;js;kt&lt;/sub&gt; which gives the probabilty that given an observable event j and (unobservable) hidden state s, the next event will be k and the hidden state t.  &lt;br&gt;&lt;br&gt;. Infering a HMM from traing sequences is more diffocult than training a plain Markov model as the hidden state isn not avaiable in the trainng data and so must itself be {{ftted}} as part of the training process. &lt;br&gt;&lt;br.&gt; Note that {{higher order Markiv modles}} (those that take into account several past observable events), which are easier to learns, can always be represented as HMMs by simply taking the hidden state to be the past tokens.  This can be one way the {{bootstrap}} the HMM training process.
)
  --></li>
  <li ><span class="name"><a href="https://alandix.com/glossary/aibook/hidden state">hidden state</a></span><!--  Array
(
    [slug] =&gt; hidden state
    [parent] =&gt; 
    [name] =&gt; hidden state
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; 
)
  --></li>
  <li ><span class="name"><a href="https://alandix.com/glossary/aibook/hierarchical grammars">hierarchical grammars</a></span><!--  Array
(
    [slug] =&gt; hierarchical grammars
    [parent] =&gt; 
    [name] =&gt; hierarchical grammars
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; 
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/aibook/high-level programming languages">high-level programming languages</a></span><!--  Array
(
    [slug] =&gt; high-level programming languages
    [parent] =&gt; 
    [name] =&gt; high-level programming languages
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; The term high-level programming language is deisgned to be closer the the human problem solution than an assembly language or similar that are operate very close to the computers raw way of dealing with values and code.  High levels languages usually include constructs for loops, functions and objects as well as ways to manipulate individual values and arrays; common examples include {{Python}} and {{Java}}.  However, there are other kinds of high-level programming language including {{logic programong}} such as {{Prolog}} or {{fucntional languages}}, such as {{Lisp}}. 
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/aibook/higher-order function">higher-order function</a></span><!--  Array
(
    [slug] =&gt; higher-order function
    [parent] =&gt; 
    [name] =&gt; higher-order function
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; A higher-order function is a function where one or more if its parameters is itself a function.  The {{map}} and {{reduce}} funcions that were the inspration for {{MapReduce}} are both examples of higher order functions.
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/aibook/hill climbing algorithm">hill climbing algorithm</a></span><!--  Array
(
    [slug] =&gt; hill climbing algorithm
    [parent] =&gt; 
    [name] =&gt; hill climbing algorithm
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; Hill climbing is one of the simplest {{optimisation}} algorithms.  On a {{search graph}} one starts at a state, and then looks at neighbouring states choosing a better one and then shifting focus to that state.  This step-by-step process is repeated until there are no neighbouring states that are better, at which point one has reached a {{local maximum}}, but not necessarily the {{global maximum}}.   The simplest form of hill climb always chooses the very best neighbouring state, but if there are many neighbouring states, or when the search is over a continuous search space, it may be necessary to choose any good-enough next step that improves on the current state.  &lt;br&gt;&lt;br&gt; Note the name 'hill climbing' suggests that the optimal solutuon is the one with highest score, but there are also many applications where the smallest score is requred, for example, distance to a target state.  In these case it is more like 'valley seeking', but the term 'hill climbing' is still applied.  Note too that in some searches, particularly {{tree search}} some states are not goal states in themselves and so do not have a score or fitness as such, but instead have a {{heuristic}} value.  In these cases one always moves to the best (or good enough) neighbour, even if this is actually worse than the heuristic of the current state.
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/aibook/hill climbing with backtracking">hill climbing with backtracking</a></span><!--  Array
(
    [slug] =&gt; hill climbing with backtracking
    [parent] =&gt; 
    [name] =&gt; hill climbing with backtracking
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; Simple {{forgetful hill climbing}} constantly tries to look foer neighbouring places or states that are better than the surrent state, but keeps no track of where it has been.   By adding {{backtracking}} hill climbing can deal impasses, when it cannot find suitable next states, or with {{local maxima}} by restarting from a previous state.  If the hill-climbing is {{deterministic}} (chooisng the very best next step) the algorithm also has to keep track of what choices it made in the past otherwise it would simply repeat the same steps after backtracking!  This is less important for {{stochasitic}} variants of hillclimbing (such as {{simulated annealing}}).
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/aibook/historical representation">historical representation</a></span><!--  Array
(
    [slug] =&gt; historical representation
    [parent] =&gt; 
    [name] =&gt; historical representation
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; In {{robotics}} a historical representation keeps track of the hstory of observationsn and actions.  Unlike a {{current state representation}}, it is able to re-assess past observations given new evidence, but it requres more memor and computation to periodically re-interpret the history.
)
  --></li>
  <li ><span class="name">HMM</span> - See <span class="has_content"><a href="https://alandix.com/glossary/aibook/hidden markov model">hidden Markov model</a></span><!--  Array
(
    [slug] =&gt; hmm
    [parent] =&gt; 
    [name] =&gt; HMM
    [canonical] =&gt; 
    [see] =&gt; hidden markov model
    [description] =&gt; 
)
  --></li>
  <li ><span class="name"><a href="https://alandix.com/glossary/aibook/hold out">hold out</a></span><!--  Array
(
    [slug] =&gt; hold out
    [parent] =&gt; 
    [name] =&gt; hold out
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; 
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/aibook/homeostasis">homeostasis</a></span><!--  Array
(
    [slug] =&gt; homeostasis
    [parent] =&gt; 
    [name] =&gt; homeostasis
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; Homeostasis is the tendency of living things to mantain a healthy or desired  state.  This applies to internal state such as body temperarure --  humans shiver to generate heat through muscle movement when too cold; and dogs pant to lose heat through evaporation when too hot.  It also applies to external state such as the small balancing movements we make when standing.  Similarly {{robots}} and forms of {{artificial life}} need to maintain 'healthy' state, such as not overheating or running out of energy.
)
  --></li>
  <li ><span class="name"><a href="https://alandix.com/glossary/aibook/homogeneous events">homogeneous events</a></span><!--  Array
(
    [slug] =&gt; homogeneous events
    [parent] =&gt; 
    [name] =&gt; homogeneous events
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; 
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/aibook/hopfield networks">Hopfield networks</a></span><!--  Array
(
    [slug] =&gt; hopfield networks
    [parent] =&gt; 
    [name] =&gt; Hopfield networks
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; A Hopfield network is a totally connected {{neural network}} often used as {{autoencoder}}.  Like {{Boltzmann machines}}, the Hopfield network is related to {[spin glass models}} in physics.  Hopfield networks differ from Boltzmann machines principally in the update model for node weights, notably, while Hopfield networks adopt a continuous learning mechanism with smooth {{thresholds}}, similar to that in {{backpropogation}}; {{Boltzmann machines}} use {{stochastic}} firing that effectively creates a {{Markov process}}, and are more similar to {{spiking neural networks}}.
)
  --></li>
  <li ><span class="name"><a href="https://alandix.com/glossary/aibook/human computation">human computation</a></span><!--  Array
(
    [slug] =&gt; human computation
    [parent] =&gt; 
    [name] =&gt; human computation
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; 
)
  --></li>
  <li ><span class="name">human in the loop</span> - See <span class="has_content"><a href="https://alandix.com/glossary/aibook/human-in-the-loop">human-in-the-loop</a></span><!--  Array
(
    [slug] =&gt; human in the loop
    [parent] =&gt; 
    [name] =&gt; human in the loop
    [canonical] =&gt; 
    [see] =&gt; human-in-the-loop
    [description] =&gt; 
)
  --></li>
  <li ><span class="name"><a href="https://alandix.com/glossary/aibook/human brain project">human brain project</a></span><!--  Array
(
    [slug] =&gt; human brain project
    [parent] =&gt; 
    [name] =&gt; human brain project
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; 
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/aibook/human-computer interaction">Human&nbsp;Computer Interaction</a></span><!--  Array
(
    [slug] =&gt; human-computer interaction
    [parent] =&gt; 
    [name] =&gt; Human&amp;nbsp;Computer Interaction
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; Human--Computer Interaction (HCI) is the study of the way people interact with technology.  It is both an academic area and also a design discipline that includes {{user experince design}} and {{user interace design}}.  Within AI it may be an application area, in particular {{intelligent user interfaces}}.  More generally whenevr an AI system is used by people there needs to be effective design adopting both general {{HCI}} principles, but also special design issues for human--AI interactions.
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/aibook/human-in-the-loop">human-in-the-loop</a></span><!--  Array
(
    [slug] =&gt; human-in-the-loop
    [parent] =&gt; 
    [name] =&gt; human-in-the-loop
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; A human-in-the-loop system si one where a human makes or strongly influences the final decsion.  For example, in some {{safety-ciritical}} situations, such as medicine or air traffic control, there may be professional or regulatory reasons way a human must oversee a final decision.  It is particularly important in {{autonomous weapons}}, where many argue that the final decison to fire shoudl come from a human.  Sometimes the term is applied more genrallly when the AI includes human participants, for exmaple online AI-driven chat whcuh may pass more complex queries to  human, altough arguably this is more properly egarded as {{human computation}}.
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/aibook/hybrid">hybrid</a></span><!--  Array
(
    [slug] =&gt; hybrid
    [parent] =&gt; 
    [name] =&gt; hybrid
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; Hybrid or hybrid system is about things that involve different knds of components, and is encountered in several ways within AI..  {{Hybrid AI}} may refer to {{hybrid architectures}} employing different kinds of AI in a single system, or {{hybrid human-AI}} applications that employ AI closely alongside humans as a form of partnership.   It may also refer to {{hybrid problems}} that include both discrete and continuous parameters.
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/aibook/hybrid ai">hybrid AI</a></span><!--  Array
(
    [slug] =&gt; hybrid ai
    [parent] =&gt; 
    [name] =&gt; hybrid AI
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; {{Hybrid AI}} may refer to {{hybrid architectures}} employing different kinds of AI in a single system, or {{hybrid human-AI}} applications that employ AI closely alongside humans as a form of partnership.  
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/aibook/hybrid architecture">hybrid architecture</a></span><!--  Array
(
    [slug] =&gt; hybrid architecture
    [parent] =&gt; 
    [name] =&gt; hybrid architecture
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; An hybrid architecture is one where different moduels or layers use different kinds of AI technolgies.  For example, we might use a {{random forest}} to perform low-level classification (say of human activity in a care setting), and then use these classes in a hand-crafted {{rule-based system}} (say &quot;if suspected fall, alert warden&quot;).
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/aibook/hybrid expert system">hybrid expert system</a></span><!--  Array
(
    [slug] =&gt; hybrid expert system
    [parent] =&gt; 
    [name] =&gt; hybrid expert system
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; A hybrid expert system is an {{expert system}} in which the AI system and expert continue to operate alongside each other after the {{training phase}}.  For example, a medical expert system might offer advice or suggest diagnoses, which the doctor might use to inform theor decison, or possibly respond by inputing additional infornation about the pattient, which may then change the expert system's advice.  Hybrid expert systems are a form of {{hybrid human-AI} system.
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/aibook/hybrid human-AI">hybrid human&ndash;AI</a></span><!--  Array
(
    [slug] =&gt; hybrid human-AI
    [parent] =&gt; 
    [name] =&gt; hybrid human&amp;ndash;AI
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; Hybrid human--AI systems are those in whch humans and AI work closely alongside each other.  While all AI systems interact with human systems at some pount, this may be quite indirect or distant, for example, an engine monitoring system in a car or and AI system creating a new medicine.  In hybrid human--A systems this connection is more direct, for example, handing control back and forth in a semi-autonomous car, or a judge using the (controversial) {{COMPAS}} system's advice in a parole hearing.  Even within hybrid human--AI systems there can be levels of cooperation, from situations where one or other is clearly the dominant, to those where it is more like a two-way cooperation, as in some forms of {{decision-support system}}.
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/aibook/hybrid problems">hybrid problems</a></span><!--  Array
(
    [slug] =&gt; hybrid problems
    [parent] =&gt; 
    [name] =&gt; hybrid problems
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; Hybrid problems are those that involve both {{continuous}} and {{discrete}} parameters or processes.  At is simplest this may be a dataset some {{continuous valued}} features and {{discrete valued}} features.  It may also refer to underlying systems that include both physical dynamics and digital control.  In the {{formal methods}} community the term 'hybrid systems' refers to the latter.  On the whole, designing algorithms to deal with &lt;em&gt;both&lt;/em&gt; continuous and discrete aspects is harder than either on their own.
)
  --></li>
  <li ><span class="name">hybrid systems</span> - See <span class="has_content"><a href="https://alandix.com/glossary/aibook/hybrid ai">hybrid AI</a></span><!--  Array
(
    [slug] =&gt; hybrid systems
    [parent] =&gt; 
    [name] =&gt; hybrid systems
    [canonical] =&gt; 
    [see] =&gt; hybrid ai
    [description] =&gt; 
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/aibook/hypertext">hypertext</a></span><!--  Array
(
    [slug] =&gt; hypertext
    [parent] =&gt; 
    [name] =&gt; hypertext
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; Hypertext is when portions of text wihin one or more documents link to one another.  The source location is often called an 'anchor', hence the tag &quot;&amp;lt;a&amp;gt;&quot; in {{HTML}}.  The target of this can point to whole document links, as commonly found in the {{web}} or this glossary, but can also include richer connections.    For example, several early hypertext systems offered multiple outward links from a single anchor point, say, to offer different links for people with different kinds expertise (engineers or users), or different levels of knowledge (high-school or university students).  Often hypertext systems, such as the {{web}} include different media such as images or videos, and the term {{hypermedia}} may be used to distinguish those where this is a major aspect of the information.  &lt;br&gt;&lt;br&gt; Within AI, hypertexts may be a source of knowledge (as in the way Wikipedia has been mined to create the {{dbPedia}} dataset), as an application to support in an {{intelligent user interface}} (for example, tracking common paths through the system and suggestingthese to future users), or as an output when links or indeed parts of the text are autimatically generated.
)
  --></li>
  <li ><span class="name"><a href="https://alandix.com/glossary/aibook/hypothesis-driven reasoning">hypothesis-driven reasoning</a></span><!--  Array
(
    [slug] =&gt; hypothesis-driven reasoning
    [parent] =&gt; 
    [name] =&gt; hypothesis-driven reasoning
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; 
)
  --></li>
  <li ><span class="name"><a href="https://alandix.com/glossary/aibook/ibm">IBM</a></span><!--  Array
(
    [slug] =&gt; ibm
    [parent] =&gt; 
    [name] =&gt; IBM
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; 
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/aibook/ibm watson">IBM Watson</a></span><!--  Array
(
    [slug] =&gt; ibm watson
    [parent] =&gt; 
    [name] =&gt; IBM Watson
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; Watson is a computer system designed by {{IBM}} to tackle general knowledge style queries.  It combines  {{natural language processing}}, {{machine learning}} and rapid access to large quantities of text from  encyclopaedias, news feeds and other human-readable data sources.   Watson is famous for defeating human contestents in the US general knowldge TV quiz show {{Jeopardy!}} in 2010.  Since then it has been adapted for other domains including medical information. 
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/aibook/id3">ID3</a></span><!--  Array
(
    [slug] =&gt; id3
    [parent] =&gt; 
    [name] =&gt; ID3
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; ID3 (Iterative Dichotomiser 3) and its later variants including {{C4.5}} of concept learning, developed by {{Ross Quinlan}}.  ID3 builds a {{decision tree}} in a top-down fashion using {{information measures}}, typically {{entropy}}, to choose a top level decision and the inductively use the same princple for lower branches.  The simplest version of ID3 simply measures the entropy gain of the decision by counting the frequency examples of each class in the branches.  As the name suggests, basic ID3 was developed for {{binary decisions}}, but there are modifications for {{non-binary decsions}}). There are also modifications to make the trees comprehensible  (see {{expaliable AI}}), either by modifying the inductive learning algorithm itself or {{pruning}} the tree once constructed.
)
  --></li>
  <li ><span class="name"><a href="https://alandix.com/glossary/aibook/identity theft">identity theft</a></span><!--  Array
(
    [slug] =&gt; identity theft
    [parent] =&gt; 
    [name] =&gt; identity theft
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; 
)
  --></li>
  <li ><span class="name">if this then that</span> - See <span class="has_content"><a href="https://alandix.com/glossary/aibook/ifttt">IFTTT </a></span><!--  Array
(
    [slug] =&gt; if this then that
    [parent] =&gt; 
    [name] =&gt; if this then that
    [canonical] =&gt; 
    [see] =&gt; ifttt
    [description] =&gt; 
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/aibook/ifttt">IFTTT </a></span><!--  Array
(
    [slug] =&gt; ifttt
    [parent] =&gt; 
    [name] =&gt; IFTTT 
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; IFTTT (if this then that) is a commercial web-based systems that allows end-users to configure rules to fire when certain events happen and conditions are met.  For example, you could configure a rule to send you a message when a particular social media chennel mentions a key term.
)
  --></li>
  <li ><span class="name"><a href="https://alandix.com/glossary/aibook/image recognition">image recognition</a></span><!--  Array
(
    [slug] =&gt; image recognition
    [parent] =&gt; 
    [name] =&gt; image recognition
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; 
)
  --></li>
  <li ><span class="name"><a href="https://alandix.com/glossary/aibook/image thresholding">image thresholding</a></span><!--  Array
(
    [slug] =&gt; image thresholding
    [parent] =&gt; 
    [name] =&gt; image thresholding
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; 
)
  --></li>
  <li ><span class="name"><a href="https://alandix.com/glossary/aibook/image understanding">image understanding</a></span><!--  Array
(
    [slug] =&gt; image understanding
    [parent] =&gt; 
    [name] =&gt; image understanding
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; 
)
  --></li>
  <li ><span class="name"><a href="https://alandix.com/glossary/aibook/imagination">imagination</a></span><!--  Array
(
    [slug] =&gt; imagination
    [parent] =&gt; 
    [name] =&gt; imagination
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; 
)
  --></li>
  <li ><span class="name"><a href="https://alandix.com/glossary/aibook/imitation game">imitation game</a></span><!--  Array
(
    [slug] =&gt; imitation game
    [parent] =&gt; 
    [name] =&gt; imitation game
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; 
)
  --></li>
  <li ><span class="name"><a href="https://alandix.com/glossary/aibook/implicit feedback">implicit feedback</a></span><!--  Array
(
    [slug] =&gt; implicit feedback
    [parent] =&gt; 
    [name] =&gt; implicit feedback
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; 
)
  --></li>
  <li ><span class="name"><a href="https://alandix.com/glossary/aibook/incidental">incidental</a></span><!--  Array
(
    [slug] =&gt; incidental
    [parent] =&gt; 
    [name] =&gt; incidental
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; 
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/aibook/independent variable">independent variable</a></span><!--  Array
(
    [slug] =&gt; independent variable
    [parent] =&gt; 
    [name] =&gt; independent variable
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; In {{linear regression}} and other foms of modelling an {{independent variable}} is a value that you can control or is an input to the model.  In contrast to a {{dependent variable}}, which is one of the  model's outputs.  That is the model is effectively of the form: &lt;br&gt; &amp;nbsp; &amp;nbsp; &amp;nbsp; dependent variables &amp;nbsp; = &amp;nbsp;  model( parameters, independent variables ) &lt;br&gt; where one then tries to fit the parameters.
)
  --></li>
  <li ><span class="name"><a href="https://alandix.com/glossary/aibook/induction">induction</a></span><!--  Array
(
    [slug] =&gt; induction
    [parent] =&gt; 
    [name] =&gt; induction
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; 
)
  --></li>
  <li ><span class="name"><a href="https://alandix.com/glossary/aibook/inductive learning">inductive learning</a></span><!--  Array
(
    [slug] =&gt; inductive learning
    [parent] =&gt; 
    [name] =&gt; inductive learning
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; 
)
  --></li>
  <li ><span class="name"><a href="https://alandix.com/glossary/aibook/inductive reasoning">inductive reasoning</a></span><!--  Array
(
    [slug] =&gt; inductive reasoning
    [parent] =&gt; 
    [name] =&gt; inductive reasoning
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; 
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/aibook/industry 4.0">Industry 4.0</a></span><!--  Array
(
    [slug] =&gt; industry 4.0
    [parent] =&gt; 
    [name] =&gt; Industry 4.0
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; Industry 4.0, sometimes also called 'fourth Industrial revolution', refers to the way in which factories and other industrial processes increasingly use smart machines ands sensor-fileld environments to monitor and control processes.
)
  --></li>
  <li ><span class="name"><a href="https://alandix.com/glossary/aibook/inexact search">inexact search</a></span><!--  Array
(
    [slug] =&gt; inexact search
    [parent] =&gt; 
    [name] =&gt; inexact search
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; 
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/aibook/infinite impulse response">infinite impulse response</a></span><!--  Array
(
    [slug] =&gt; infinite impulse response
    [parent] =&gt; 
    [name] =&gt; infinite impulse response
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; In an infinite input response (IIR) systemthe current response may be influenced by observations or events arbitrarily far in the past. Typically this arises when ther is some form of feedback loop of past outputs of the system as part of the current input or where there is internal state.  Some statistical time series analysis techniques are IIR such as  a {{auto-regressive}} filters as well as neural network techniqus such as {{recurrent neural networks}}.
)
  --></li>
  <li ><span class="name"><a href="https://alandix.com/glossary/aibook/information">information</a></span><!--  Array
(
    [slug] =&gt; information
    [parent] =&gt; 
    [name] =&gt; information
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; 
)
  --></li>
  <li ><span class="name">information entropy</span> - See <span class="has_content"><a href="https://alandix.com/glossary/aibook/entropy">entropy</a></span><!--  Array
(
    [slug] =&gt; information entropy
    [parent] =&gt; 
    [name] =&gt; information entropy
    [canonical] =&gt; 
    [see] =&gt; entropy
    [description] =&gt; 
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/aibook/information gain">information gain</a></span><!--  Array
(
    [slug] =&gt; information gain
    [parent] =&gt; 
    [name] =&gt; information gain
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; You may be familiar with the &lt;em&gt;base&lt;/em&gt; 2  {{logarithm}} of a number N (written log&lt;sub&gt;2&lt;/sub&gt;(N), or log2(N).  This is defined to be the number L such that N = 2&lt;sup&gt;L&lt;/sup&gt;.  For example,  log&lt;sub&gt;2&lt;/sub&gt;(32) = 5 and log&lt;sub&gt;2&lt;/sub&gt;(10) &amp;tilde; 3.322.  You may also have come across logarithms base 10, so that log&lt;sub&gt;10&lt;/sub&gt;(1000) = 3.   The &lt;em&gt;base of a logarithm&lt;/em&gt; is the generalisation of this; so that if B is the base, then the logarithm v=base B of a nuymber N (written  log&lt;sub&gt;B&lt;/sub&gt;(N)) satisfies: &lt;br&gt;  &amp;nbsp;  &amp;nbsp;  &amp;nbsp;  N  &amp;nbsp; =  &amp;nbsp;  B&lt;sup&gt;log&lt;sub&gt;B&lt;/sub&gt;(N)&lt;/sup&gt; &lt;br&gt; &lt;/br&gt;As well as logarithms base 2 (common in comouting becasue of the use of the binary system) and base 10, you are lielly to come across what are known as the {{natural logarithm}} that is the logarithm to base e, where e is {{Eular's number}} &amp;tilde; 2.7182.  This doesn't sound like a very 'natural' base fo a logarithm, but they occur commonly on physical and bilogical systems. &lt;br&gt;&lt;br&gt; The choice of base only affects the scale of the locgarithm as all logarithms are linearly related to one another and in general given two logarithm bases A and B:&lt;br&gt;  &amp;nbsp;  &amp;nbsp;  &amp;nbsp;  log&lt;sub&gt;A&lt;/sub&gt;(N) &amp;nbsp; = &amp;nbsp;   log&lt;sub&gt;A&lt;/sub&gt;(B) &amp;times;  log&lt;sub&gt;B&lt;/sub&gt;(N)
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/aibook/information gain ratio">information gain ratio</a></span><!--  Array
(
    [slug] =&gt; information gain ratio
    [parent] =&gt; 
    [name] =&gt; information gain ratio
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; Information gain ratio is an {{information measure}} used in {{decision tree}} bulding algorithms, notably {{C4.5}}, that tries to remove some of the bias towards decsions with lareg numbers of alternatives.  It is obtained by dividing the raw {{information gain}} (gain(C,G)) by the entropy of the decision grouping.  That is &lt;br&gt; &amp;nbsp; Information gain ratio =  gain(C,G) / H(G) &lt;br&gt; where H(G) is the entropy of the division into the groups. &lt;br&gt;&lt;br&gt;. This is similar to but different from {{normalised mutual information}}, as the latter divides by H(C)+H(G).   In practice this means the information gain ratio it is weighted slightly more against larger numbers of groups compared with  {{normalised mutual information}}.
)
  --></li>
  <li ><span class="name"><a href="https://alandix.com/glossary/aibook/information preserving">information preserving</a></span><!--  Array
(
    [slug] =&gt; information preserving
    [parent] =&gt; 
    [name] =&gt; information preserving
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; 
)
  --></li>
  <li ><span class="name"><a href="https://alandix.com/glossary/aibook/information processing">information processing</a></span><!--  Array
(
    [slug] =&gt; information processing
    [parent] =&gt; 
    [name] =&gt; information processing
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; 
)
  --></li>
  <li ><span class="name"><a href="https://alandix.com/glossary/aibook/inhibition">inhibition</a></span><!--  Array
(
    [slug] =&gt; inhibition
    [parent] =&gt; 
    [name] =&gt; inhibition
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; 
)
  --></li>
  <li ><span class="name"><a href="https://alandix.com/glossary/aibook/insight">insight</a></span><!--  Array
(
    [slug] =&gt; insight
    [parent] =&gt; 
    [name] =&gt; insight
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; 
)
  --></li>
  <li ><span class="name"><a href="https://alandix.com/glossary/aibook/instabilities">instabilities</a></span><!--  Array
(
    [slug] =&gt; instabilities
    [parent] =&gt; 
    [name] =&gt; instabilities
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; 
)
  --></li>
  <li ><span class="name"><a href="https://alandix.com/glossary/aibook/intellectual property">intellectual property</a></span><!--  Array
(
    [slug] =&gt; intellectual property
    [parent] =&gt; 
    [name] =&gt; intellectual property
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; 
)
  --></li>
  <li ><span class="name"><a href="https://alandix.com/glossary/aibook/intelligence">intelligence</a></span>- has subentries<ul class="subentries"><li ><span class="name"><a href="https://alandix.com/glossary/aibook/intelligence!computation view">computation view</a></span><!--  Array
(
    [slug] =&gt; intelligence!computation view
    [parent] =&gt; intelligence
    [name] =&gt; computation view
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; 
)
  --></li>
<li ><span class="name"><a href="https://alandix.com/glossary/aibook/intelligence!emotion">emotion</a></span><!--  Array
(
    [slug] =&gt; intelligence!emotion
    [parent] =&gt; intelligence
    [name] =&gt; emotion
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; 
)
  --></li>
<li ><span class="name"><a href="https://alandix.com/glossary/aibook/intelligence!morals">morals</a></span><!--  Array
(
    [slug] =&gt; intelligence!morals
    [parent] =&gt; intelligence
    [name] =&gt; morals
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; 
)
  --></li>
<li ><span class="name"><a href="https://alandix.com/glossary/aibook/intelligence!responsibility">responsibility</a></span><!--  Array
(
    [slug] =&gt; intelligence!responsibility
    [parent] =&gt; intelligence
    [name] =&gt; responsibility
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; 
)
  --></li></ul><!--  Array
(
    [slug] =&gt; intelligence
    [parent] =&gt; 
    [name] =&gt; intelligence
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; 
    [subentries] =&gt; Array
        (
            [intelligence!computation view] =&gt; Array
                (
                    [slug] =&gt; intelligence!computation view
                    [parent] =&gt; intelligence
                    [name] =&gt; computation view
                    [canonical] =&gt; 
                    [see] =&gt; 
                    [description] =&gt; 
                )

            [intelligence!emotion] =&gt; Array
                (
                    [slug] =&gt; intelligence!emotion
                    [parent] =&gt; intelligence
                    [name] =&gt; emotion
                    [canonical] =&gt; 
                    [see] =&gt; 
                    [description] =&gt; 
                )

            [intelligence!morals] =&gt; Array
                (
                    [slug] =&gt; intelligence!morals
                    [parent] =&gt; intelligence
                    [name] =&gt; morals
                    [canonical] =&gt; 
                    [see] =&gt; 
                    [description] =&gt; 
                )

            [intelligence!responsibility] =&gt; Array
                (
                    [slug] =&gt; intelligence!responsibility
                    [parent] =&gt; intelligence
                    [name] =&gt; responsibility
                    [canonical] =&gt; 
                    [see] =&gt; 
                    [description] =&gt; 
                )

        )

)
  --></li>
  <li ><span class="name"><a href="https://alandix.com/glossary/aibook/intelligent user interfaces">intelligent user interfaces</a></span><!--  Array
(
    [slug] =&gt; intelligent user interfaces
    [parent] =&gt; 
    [name] =&gt; intelligent user interfaces
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; 
)
  --></li>
  <li ><span class="name"><a href="https://alandix.com/glossary/aibook/intent">intent</a></span><!--  Array
(
    [slug] =&gt; intent
    [parent] =&gt; 
    [name] =&gt; intent
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; 
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/aibook/intentional representation">intentional representation</a></span><!--  Array
(
    [slug] =&gt; intentional representation
    [parent] =&gt; 
    [name] =&gt; intentional representation
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; For a {{logical expression}} or {{database query}} the intensional representation is the formulae of the expression or the query expresson in a notation such as {{SQL}}.  In constrats the  {{extensional representation}} is the collection of all values that satisfy the expression or are returned by the query.
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/aibook/intentionality">intentionality</a></span><!--  Array
(
    [slug] =&gt; intentionality
    [parent] =&gt; 
    [name] =&gt; intentionality
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; In the philosophy of intelligence and consciousness, intentionality refers to the way that words are not merely sounds, but have real menaing.  When a person says &quot;&quot;the sky is blue&quot;&quot;, it is nt just a statement that has been garnered from weather data or observed through the window, but the idea of 'sky' an ''blueness' are attached to a real sense of what the person means by 'blue'
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/aibook/interactive voice response">interactive voice response</a></span><!--  Array
(
    [slug] =&gt; interactive voice response
    [parent] =&gt; 
    [name] =&gt; interactive voice response
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; Interactive voice response (IVR) systems are telephone-based interactive user interfaces which combine voicxe output form the computer with user input in the form of keypresses on the phone or simple human voice responses such as &quot;Yes/No&quot; or speaking digits.  Most systems follow a fixed dialgogue tree, where the user responses determine which path to follow.
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/aibook/intercept">intercept</a></span><!--  Array
(
    [slug] =&gt; intercept
    [parent] =&gt; 
    [name] =&gt; intercept
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; When drawing a straight line and especially in {{linear regression}} o an x--y graph, the intercept is the place where the line crosses the y axis, in other words the value when x=0.. In the case when there is a single {{independent variable}} (say time) and many {{dependent variables}}, the intercept is a vector.
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/aibook/internet of things">internet of things</a></span><!--  Array
(
    [slug] =&gt; internet of things
    [parent] =&gt; 
    [name] =&gt; internet of things
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; The internet of things (IoT) refers to the way in which many everyday devices are now internet-enabled and thus part of the global network.  This includes domestic devices (e.g. internet-enabled kettles, smart-fridges, and door-step cameras), {{smart cities}} (e.g. traffic lights, pedestrian flow sensors, pollution monitoring) and industrial applications (e.g. stock monitoring through {{NFC}}, industrial robots).  IoT offers the oppoortunity to interlink these devices and thus allow applications, such as automatically adding milk to your internet shopping list when it runs low in the fridge, or adjusting traffic light timings during epriods of conjestion.  However, there are worries about {{privacy}}  (e.g. if the supermarket learns about your daily eating) and {{security}} (e.g. when building heating systems have been targetted by cybercriminals). 
)
  --></li>
  <li ><span class="name"><a href="https://alandix.com/glossary/aibook/interpret">interpret</a></span><!--  Array
(
    [slug] =&gt; interpret
    [parent] =&gt; 
    [name] =&gt; interpret
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; 
)
  --></li>
  <li ><span class="name"><a href="https://alandix.com/glossary/aibook/interpreter">interpreter</a></span><!--  Array
(
    [slug] =&gt; interpreter
    [parent] =&gt; 
    [name] =&gt; interpreter
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; 
)
  --></li>
  <li ><span class="name">IoT</span> - See <span class="has_content"><a href="https://alandix.com/glossary/aibook/internet of things">internet of things</a></span><!--  Array
(
    [slug] =&gt; iot
    [parent] =&gt; 
    [name] =&gt; IoT
    [canonical] =&gt; 
    [see] =&gt; internet of things
    [description] =&gt; 
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/aibook/isbn">ISBN</a></span><!--  Array
(
    [slug] =&gt; isbn
    [parent] =&gt; 
    [name] =&gt; ISBN
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; An ISBN (International Book Serial Number)is a {{unique identifier}} for a version of a book.   Different editions and paperback and hardback versions of the same book have different ISBNs. 
)
  --></li>
  <li ><span class="name"><a href="https://alandix.com/glossary/aibook/iterative broadening">iterative broadening</a></span><!--  Array
(
    [slug] =&gt; iterative broadening
    [parent] =&gt; 
    [name] =&gt; iterative broadening
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; 
)
  --></li>
  <li ><span class="name"><a href="https://alandix.com/glossary/aibook/iterative deepening">iterative deepening</a></span><!--  Array
(
    [slug] =&gt; iterative deepening
    [parent] =&gt; 
    [name] =&gt; iterative deepening
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; 
)
  --></li>
  <li ><span class="name">IVR</span> - See <span class="has_content"><a href="https://alandix.com/glossary/aibook/interactive voice response">interactive voice response</a></span><!--  Array
(
    [slug] =&gt; ivr
    [parent] =&gt; 
    [name] =&gt; IVR
    [canonical] =&gt; 
    [see] =&gt; interactive voice response
    [description] =&gt; 
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/aibook/jaccard similarity">Jaccard similarity</a></span><!--  Array
(
    [slug] =&gt; jaccard similarity
    [parent] =&gt; 
    [name] =&gt; Jaccard similarity
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; Jaccard similarity is a measure of {{similarity}} between two doucuments.  The Jaccard similarity to compare two documents, doc&lt;sub&gt;1&lt;/sub&gt; and doc&lt;sub&gt;2&lt;/sub&gt;, works on the {{bag of words}} (say words&lt;sub&gt;1&lt;/sub&gt; and words&lt;sub&gt;2&lt;/sub&gt;, and then calculates &lt;br&gt; &amp;nbsp;&amp;nbsp;&amp;nbsp;&amp;nbsp; | words&lt;sub&gt;1&lt;/sub&gt; &amp;cap; words&lt;sub&gt;2&lt;/sub&gt; | / | words&lt;sub&gt;1&lt;/sub&gt; &amp;cup; words&lt;sub&gt;2&lt;/sub&gt; | &lt;br&gt;. That is the number of distinct words in both documents divided by the number of distinct words in the union.  The Jaccard similarity is used heavily in {{document retrieval}} algorithms.
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/aibook/jacquard loom">Jacquard loom</a></span><!--  Array
(
    [slug] =&gt; jacquard loom
    [parent] =&gt; 
    [name] =&gt; Jacquard loom
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; The Jacquard loom used punch cards to allow a rich variety of weavng patterns, far beyond fixed methods of tying up that preceded it.  It was the inspiration for the punch cards used in {{Babbage}}'s {{Analytical Engine}} and hence the Jacquard loom can be seen as a direct desc endent of the modern programmable computer. Indeed, these orgins inspired {{Ada Lovelace}} to write about the way the {{Analytical Engine}} &quot;&quot;weaves algebraical patterns just as the Jacquard-loom weaves flowers and leaves&quot;.
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/aibook/jeopardy">Jeopardy!</a></span><!--  Array
(
    [slug] =&gt; jeopardy
    [parent] =&gt; 
    [name] =&gt; Jeopardy!
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; Jeopardy! is US general knowldge TV quiz show.  {{IBM Watson}} computer won the game in 2010 against human openents.  It used a combination of natural language processing, machine learning and rapid access to large quantities of text from  encyclopaedias, news feeds and other human-readable data sources.
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/aibook/junk dna">junk DNA</a></span><!--  Array
(
    [slug] =&gt; junk dna
    [parent] =&gt; 
    [name] =&gt; junk DNA
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; The human chromosome contains large tracts of DNA which does not appear to code for anything useful.  This is known as junk DNA.
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/aibook/k-fold cross-validation">k-fold cross-validation</a></span><!--  Array
(
    [slug] =&gt; k-fold cross-validation
    [parent] =&gt; 
    [name] =&gt; k-fold cross-validation
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; K-fold cross-validation is a technique used to validate a {{machine learning}} algorithm for a particular dataset.  The dataset is divided into k pieces (of size N/k, where n is the size of the full dataset) and then the algorithm is trained on k-1 of them with the remaining N/k of the dataset used as test data for an {{accuracy measure}}.  This is then repeated with a different piece left out and the results of the {{accuracy measure}} averaged.
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/aibook/k-means algorithm">k-means algorithm</a></span><!--  Array
(
    [slug] =&gt; k-means algorithm
    [parent] =&gt; 
    [name] =&gt; k-means algorithm
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; K-means is one of the oldest commonly used {{clustering}} algorithms.
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/aibook/kasparov, garry">Kasparov, Garry</a></span><!--  Array
(
    [slug] =&gt; kasparov, garry
    [parent] =&gt; 
    [name] =&gt; Kasparov, Garry
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; Garry Kasparov is a chess grandmaster and world champion for many years, who was famously defeated by the IBM chess programme {{Deep Blue}} in 1996.
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/aibook/kernel">kernel</a></span><!--  Array
(
    [slug] =&gt; kernel
    [parent] =&gt; 
    [name] =&gt; kernel
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; Kernel functions are used in {{machine learning}} including {{support vector machines}} to increas {{non-linear diversity}}.  The term originates in mathematics where a postive definite kernel is a form of {{non-linear}} generalisation of the {{inner prooduct}}.
)
  --></li>
  <li ><span class="name"><a href="https://alandix.com/glossary/aibook/knowledge acquisition">knowledge acquisition</a></span><!--  Array
(
    [slug] =&gt; knowledge acquisition
    [parent] =&gt; 
    [name] =&gt; knowledge acquisition
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; 
)
  --></li>
  <li ><span class="name"><a href="https://alandix.com/glossary/aibook/knowledge base">knowledge base</a></span><!--  Array
(
    [slug] =&gt; knowledge base
    [parent] =&gt; 
    [name] =&gt; knowledge base
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; 
)
  --></li>
  <li ><span class="name"><a href="https://alandix.com/glossary/aibook/knowledge elicitation">knowledge elicitation</a></span>- has subentries<ul class="subentries"><li ><span class="name"><a href="https://alandix.com/glossary/aibook/knowledge elicitation!critiquing">critiquing</a></span><!--  Array
(
    [slug] =&gt; knowledge elicitation!critiquing
    [parent] =&gt; knowledge elicitation
    [name] =&gt; critiquing
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; 
)
  --></li>
<li ><span class="name"><a href="https://alandix.com/glossary/aibook/knowledge elicitation!focused discussions">focused discussions</a></span><!--  Array
(
    [slug] =&gt; knowledge elicitation!focused discussions
    [parent] =&gt; knowledge elicitation
    [name] =&gt; focused discussions
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; 
)
  --></li>
<li ><span class="name"><a href="https://alandix.com/glossary/aibook/knowledge elicitation!post-task walkthrough">post-task walkthrough</a></span><!--  Array
(
    [slug] =&gt; knowledge elicitation!post-task walkthrough
    [parent] =&gt; knowledge elicitation
    [name] =&gt; post-task walkthrough
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; 
)
  --></li>
<li ><span class="name"><a href="https://alandix.com/glossary/aibook/knowledge elicitation!role reversal">role reversal</a></span><!--  Array
(
    [slug] =&gt; knowledge elicitation!role reversal
    [parent] =&gt; knowledge elicitation
    [name] =&gt; role reversal
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; 
)
  --></li>
<li ><span class="name"><a href="https://alandix.com/glossary/aibook/knowledge elicitation!seed questions">seed questions</a></span><!--  Array
(
    [slug] =&gt; knowledge elicitation!seed questions
    [parent] =&gt; knowledge elicitation
    [name] =&gt; seed questions
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; 
)
  --></li>
<li ><span class="name"><a href="https://alandix.com/glossary/aibook/knowledge elicitation!structured interview">structured interview</a></span><!--  Array
(
    [slug] =&gt; knowledge elicitation!structured interview
    [parent] =&gt; knowledge elicitation
    [name] =&gt; structured interview
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; 
)
  --></li>
<li ><span class="name"><a href="https://alandix.com/glossary/aibook/knowledge elicitation!teach-back interview">teach-back interview</a></span><!--  Array
(
    [slug] =&gt; knowledge elicitation!teach-back interview
    [parent] =&gt; knowledge elicitation
    [name] =&gt; teach-back interview
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; 
)
  --></li>
<li ><span class="name"><a href="https://alandix.com/glossary/aibook/knowledge elicitation!think aloud">think aloud</a></span><!--  Array
(
    [slug] =&gt; knowledge elicitation!think aloud
    [parent] =&gt; knowledge elicitation
    [name] =&gt; think aloud
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; 
)
  --></li>
<li ><span class="name"><a href="https://alandix.com/glossary/aibook/knowledge elicitation!twenty questions">twenty questions</a></span><!--  Array
(
    [slug] =&gt; knowledge elicitation!twenty questions
    [parent] =&gt; knowledge elicitation
    [name] =&gt; twenty questions
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; 
)
  --></li>
<li ><span class="name"><a href="https://alandix.com/glossary/aibook/knowledge elicitation!unstructured interview">unstructured interview</a></span><!--  Array
(
    [slug] =&gt; knowledge elicitation!unstructured interview
    [parent] =&gt; knowledge elicitation
    [name] =&gt; unstructured interview
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; 
)
  --></li></ul><!--  Array
(
    [slug] =&gt; knowledge elicitation
    [parent] =&gt; 
    [name] =&gt; knowledge elicitation
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; 
    [subentries] =&gt; Array
        (
            [knowledge elicitation!critiquing] =&gt; Array
                (
                    [slug] =&gt; knowledge elicitation!critiquing
                    [parent] =&gt; knowledge elicitation
                    [name] =&gt; critiquing
                    [canonical] =&gt; 
                    [see] =&gt; 
                    [description] =&gt; 
                )

            [knowledge elicitation!focused discussions] =&gt; Array
                (
                    [slug] =&gt; knowledge elicitation!focused discussions
                    [parent] =&gt; knowledge elicitation
                    [name] =&gt; focused discussions
                    [canonical] =&gt; 
                    [see] =&gt; 
                    [description] =&gt; 
                )

            [knowledge elicitation!post-task walkthrough] =&gt; Array
                (
                    [slug] =&gt; knowledge elicitation!post-task walkthrough
                    [parent] =&gt; knowledge elicitation
                    [name] =&gt; post-task walkthrough
                    [canonical] =&gt; 
                    [see] =&gt; 
                    [description] =&gt; 
                )

            [knowledge elicitation!role reversal] =&gt; Array
                (
                    [slug] =&gt; knowledge elicitation!role reversal
                    [parent] =&gt; knowledge elicitation
                    [name] =&gt; role reversal
                    [canonical] =&gt; 
                    [see] =&gt; 
                    [description] =&gt; 
                )

            [knowledge elicitation!seed questions] =&gt; Array
                (
                    [slug] =&gt; knowledge elicitation!seed questions
                    [parent] =&gt; knowledge elicitation
                    [name] =&gt; seed questions
                    [canonical] =&gt; 
                    [see] =&gt; 
                    [description] =&gt; 
                )

            [knowledge elicitation!structured interview] =&gt; Array
                (
                    [slug] =&gt; knowledge elicitation!structured interview
                    [parent] =&gt; knowledge elicitation
                    [name] =&gt; structured interview
                    [canonical] =&gt; 
                    [see] =&gt; 
                    [description] =&gt; 
                )

            [knowledge elicitation!teach-back interview] =&gt; Array
                (
                    [slug] =&gt; knowledge elicitation!teach-back interview
                    [parent] =&gt; knowledge elicitation
                    [name] =&gt; teach-back interview
                    [canonical] =&gt; 
                    [see] =&gt; 
                    [description] =&gt; 
                )

            [knowledge elicitation!think aloud] =&gt; Array
                (
                    [slug] =&gt; knowledge elicitation!think aloud
                    [parent] =&gt; knowledge elicitation
                    [name] =&gt; think aloud
                    [canonical] =&gt; 
                    [see] =&gt; 
                    [description] =&gt; 
                )

            [knowledge elicitation!twenty questions] =&gt; Array
                (
                    [slug] =&gt; knowledge elicitation!twenty questions
                    [parent] =&gt; knowledge elicitation
                    [name] =&gt; twenty questions
                    [canonical] =&gt; 
                    [see] =&gt; 
                    [description] =&gt; 
                )

            [knowledge elicitation!unstructured interview] =&gt; Array
                (
                    [slug] =&gt; knowledge elicitation!unstructured interview
                    [parent] =&gt; knowledge elicitation
                    [name] =&gt; unstructured interview
                    [canonical] =&gt; 
                    [see] =&gt; 
                    [description] =&gt; 
                )

        )

)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/aibook/knowledge engineer">knowledge engineer</a></span><!--  Array
(
    [slug] =&gt; knowledge engineer
    [parent] =&gt; 
    [name] =&gt; knowledge engineer
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; A knowledeg engoneer is a professional who works with a {{domain expert}} to elicit and represent the expert's knowledge for use in an {{expert system}} or other form of AI.
)
  --></li>
  <li ><span class="name"><a href="https://alandix.com/glossary/aibook/knowledge">knowledge</a></span>- has subentries<ul class="subentries"><li ><span class="name">declarative</span> - See <span class="has_content"><a href="https://alandix.com/glossary/aibook/declarative knowledge">declarative knowledge</a></span><!--  Array
(
    [slug] =&gt; knowledge!declarative
    [parent] =&gt; knowledge
    [name] =&gt; declarative
    [canonical] =&gt; 
    [see] =&gt; declarative knowledge
    [description] =&gt; 
)
  --></li>
<li ><span class="name">domain-independent</span> - See <span class="has_content"><a href="https://alandix.com/glossary/aibook/domain-independent knowledge">domain-independent knowledge</a></span><!--  Array
(
    [slug] =&gt; knowledge!domain-independent
    [parent] =&gt; knowledge
    [name] =&gt; domain-independent
    [canonical] =&gt; 
    [see] =&gt; domain-independent knowledge
    [description] =&gt; 
)
  --></li>
<li ><span class="name">domain-specific</span> - See <span class="has_content"><a href="https://alandix.com/glossary/aibook/domain-specific knowledge">domain-specific knowledge</a></span><!--  Array
(
    [slug] =&gt; knowledge!domain-specific
    [parent] =&gt; knowledge
    [name] =&gt; domain-specific
    [canonical] =&gt; 
    [see] =&gt; domain-specific knowledge
    [description] =&gt; 
)
  --></li>
<li ><span class="name">procedural</span> - See <span class="has_content"><a href="https://alandix.com/glossary/aibook/procedural knowledge">procedural knowledge</a></span><!--  Array
(
    [slug] =&gt; knowledge!procedural
    [parent] =&gt; knowledge
    [name] =&gt; procedural
    [canonical] =&gt; 
    [see] =&gt; procedural knowledge
    [description] =&gt; 
)
  --></li></ul><!--  Array
(
    [slug] =&gt; knowledge
    [parent] =&gt; 
    [name] =&gt; knowledge
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; 
    [subentries] =&gt; Array
        (
            [knowledge!declarative] =&gt; Array
                (
                    [slug] =&gt; knowledge!declarative
                    [parent] =&gt; knowledge
                    [name] =&gt; declarative
                    [canonical] =&gt; 
                    [see] =&gt; declarative knowledge
                    [description] =&gt; 
                )

            [knowledge!domain-independent] =&gt; Array
                (
                    [slug] =&gt; knowledge!domain-independent
                    [parent] =&gt; knowledge
                    [name] =&gt; domain-independent
                    [canonical] =&gt; 
                    [see] =&gt; domain-independent knowledge
                    [description] =&gt; 
                )

            [knowledge!domain-specific] =&gt; Array
                (
                    [slug] =&gt; knowledge!domain-specific
                    [parent] =&gt; knowledge
                    [name] =&gt; domain-specific
                    [canonical] =&gt; 
                    [see] =&gt; domain-specific knowledge
                    [description] =&gt; 
                )

            [knowledge!procedural] =&gt; Array
                (
                    [slug] =&gt; knowledge!procedural
                    [parent] =&gt; knowledge
                    [name] =&gt; procedural
                    [canonical] =&gt; 
                    [see] =&gt; procedural knowledge
                    [description] =&gt; 
                )

        )

)
  --></li>
  <li ><span class="name"><a href="https://alandix.com/glossary/aibook/knowledge representation">knowledge representation</a></span>- has subentries<ul class="subentries"><li ><span class="name"><a href="https://alandix.com/glossary/aibook/knowledge representation!by logic">by logic</a></span><!--  Array
(
    [slug] =&gt; knowledge representation!by logic
    [parent] =&gt; knowledge representation
    [name] =&gt; by logic
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; 
)
  --></li>
<li ><span class="name"><a href="https://alandix.com/glossary/aibook/knowledge representation!by networks">by networks</a></span><!--  Array
(
    [slug] =&gt; knowledge representation!by networks
    [parent] =&gt; knowledge representation
    [name] =&gt; by networks
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; 
)
  --></li>
<li ><span class="name"><a href="https://alandix.com/glossary/aibook/knowledge representation!clarity">clarity</a></span><!--  Array
(
    [slug] =&gt; knowledge representation!clarity
    [parent] =&gt; knowledge representation
    [name] =&gt; clarity
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; 
)
  --></li>
<li ><span class="name"><a href="https://alandix.com/glossary/aibook/knowledge representation!completeness">completeness</a></span><!--  Array
(
    [slug] =&gt; knowledge representation!completeness
    [parent] =&gt; knowledge representation
    [name] =&gt; completeness
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; 
)
  --></li>
<li ><span class="name"><a href="https://alandix.com/glossary/aibook/knowledge representation!effectiveness">effectiveness</a></span><!--  Array
(
    [slug] =&gt; knowledge representation!effectiveness
    [parent] =&gt; knowledge representation
    [name] =&gt; effectiveness
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; 
)
  --></li>
<li ><span class="name"><a href="https://alandix.com/glossary/aibook/knowledge representation!efficiency">efficiency</a></span><!--  Array
(
    [slug] =&gt; knowledge representation!efficiency
    [parent] =&gt; knowledge representation
    [name] =&gt; efficiency
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; 
)
  --></li>
<li ><span class="name"><a href="https://alandix.com/glossary/aibook/knowledge representation!explicitness">explicitness</a></span><!--  Array
(
    [slug] =&gt; knowledge representation!explicitness
    [parent] =&gt; knowledge representation
    [name] =&gt; explicitness
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; 
)
  --></li>
<li ><span class="name"><a href="https://alandix.com/glossary/aibook/knowledge representation!expressiveness">expressiveness</a></span><!--  Array
(
    [slug] =&gt; knowledge representation!expressiveness
    [parent] =&gt; knowledge representation
    [name] =&gt; expressiveness
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; 
)
  --></li>
<li ><span class="name"><a href="https://alandix.com/glossary/aibook/knowledge representation!frame">frame</a></span><!--  Array
(
    [slug] =&gt; knowledge representation!frame
    [parent] =&gt; knowledge representation
    [name] =&gt; frame
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; 
)
  --></li>
<li ><span class="name"><a href="https://alandix.com/glossary/aibook/knowledge representation!granularity">granularity</a></span><!--  Array
(
    [slug] =&gt; knowledge representation!granularity
    [parent] =&gt; knowledge representation
    [name] =&gt; granularity
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; 
)
  --></li>
<li ><span class="name"><a href="https://alandix.com/glossary/aibook/knowledge representation!metrics">metrics</a></span><!--  Array
(
    [slug] =&gt; knowledge representation!metrics
    [parent] =&gt; knowledge representation
    [name] =&gt; metrics
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; 
)
  --></li>
<li ><span class="name">procedural</span> - See <span class="has_content"><a href="https://alandix.com/glossary/aibook/procedural knowledge">procedural knowledge</a></span><!--  Array
(
    [slug] =&gt; knowledge representation!procedural
    [parent] =&gt; knowledge representation
    [name] =&gt; procedural
    [canonical] =&gt; 
    [see] =&gt; procedural knowledge
    [description] =&gt; 
)
  --></li>
<li ><span class="name"><a href="https://alandix.com/glossary/aibook/knowledge representation!script">script</a></span><!--  Array
(
    [slug] =&gt; knowledge representation!script
    [parent] =&gt; knowledge representation
    [name] =&gt; script
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; 
)
  --></li>
<li ><span class="name"><a href="https://alandix.com/glossary/aibook/knowledge representation!structured">structured</a></span><!--  Array
(
    [slug] =&gt; knowledge representation!structured
    [parent] =&gt; knowledge representation
    [name] =&gt; structured
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; 
)
  --></li></ul><!--  Array
(
    [slug] =&gt; knowledge representation
    [parent] =&gt; 
    [name] =&gt; knowledge representation
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; 
    [subentries] =&gt; Array
        (
            [knowledge representation!by logic] =&gt; Array
                (
                    [slug] =&gt; knowledge representation!by logic
                    [parent] =&gt; knowledge representation
                    [name] =&gt; by logic
                    [canonical] =&gt; 
                    [see] =&gt; 
                    [description] =&gt; 
                )

            [knowledge representation!by networks] =&gt; Array
                (
                    [slug] =&gt; knowledge representation!by networks
                    [parent] =&gt; knowledge representation
                    [name] =&gt; by networks
                    [canonical] =&gt; 
                    [see] =&gt; 
                    [description] =&gt; 
                )

            [knowledge representation!clarity] =&gt; Array
                (
                    [slug] =&gt; knowledge representation!clarity
                    [parent] =&gt; knowledge representation
                    [name] =&gt; clarity
                    [canonical] =&gt; 
                    [see] =&gt; 
                    [description] =&gt; 
                )

            [knowledge representation!completeness] =&gt; Array
                (
                    [slug] =&gt; knowledge representation!completeness
                    [parent] =&gt; knowledge representation
                    [name] =&gt; completeness
                    [canonical] =&gt; 
                    [see] =&gt; 
                    [description] =&gt; 
                )

            [knowledge representation!effectiveness] =&gt; Array
                (
                    [slug] =&gt; knowledge representation!effectiveness
                    [parent] =&gt; knowledge representation
                    [name] =&gt; effectiveness
                    [canonical] =&gt; 
                    [see] =&gt; 
                    [description] =&gt; 
                )

            [knowledge representation!efficiency] =&gt; Array
                (
                    [slug] =&gt; knowledge representation!efficiency
                    [parent] =&gt; knowledge representation
                    [name] =&gt; efficiency
                    [canonical] =&gt; 
                    [see] =&gt; 
                    [description] =&gt; 
                )

            [knowledge representation!explicitness] =&gt; Array
                (
                    [slug] =&gt; knowledge representation!explicitness
                    [parent] =&gt; knowledge representation
                    [name] =&gt; explicitness
                    [canonical] =&gt; 
                    [see] =&gt; 
                    [description] =&gt; 
                )

            [knowledge representation!expressiveness] =&gt; Array
                (
                    [slug] =&gt; knowledge representation!expressiveness
                    [parent] =&gt; knowledge representation
                    [name] =&gt; expressiveness
                    [canonical] =&gt; 
                    [see] =&gt; 
                    [description] =&gt; 
                )

            [knowledge representation!frame] =&gt; Array
                (
                    [slug] =&gt; knowledge representation!frame
                    [parent] =&gt; knowledge representation
                    [name] =&gt; frame
                    [canonical] =&gt; 
                    [see] =&gt; 
                    [description] =&gt; 
                )

            [knowledge representation!granularity] =&gt; Array
                (
                    [slug] =&gt; knowledge representation!granularity
                    [parent] =&gt; knowledge representation
                    [name] =&gt; granularity
                    [canonical] =&gt; 
                    [see] =&gt; 
                    [description] =&gt; 
                )

            [knowledge representation!metrics] =&gt; Array
                (
                    [slug] =&gt; knowledge representation!metrics
                    [parent] =&gt; knowledge representation
                    [name] =&gt; metrics
                    [canonical] =&gt; 
                    [see] =&gt; 
                    [description] =&gt; 
                )

            [knowledge representation!procedural] =&gt; Array
                (
                    [slug] =&gt; knowledge representation!procedural
                    [parent] =&gt; knowledge representation
                    [name] =&gt; procedural
                    [canonical] =&gt; 
                    [see] =&gt; procedural knowledge
                    [description] =&gt; 
                )

            [knowledge representation!script] =&gt; Array
                (
                    [slug] =&gt; knowledge representation!script
                    [parent] =&gt; knowledge representation
                    [name] =&gt; script
                    [canonical] =&gt; 
                    [see] =&gt; 
                    [description] =&gt; 
                )

            [knowledge representation!structured] =&gt; Array
                (
                    [slug] =&gt; knowledge representation!structured
                    [parent] =&gt; knowledge representation
                    [name] =&gt; structured
                    [canonical] =&gt; 
                    [see] =&gt; 
                    [description] =&gt; 
                )

        )

)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/aibook/knowledge-rich search">knowledge-rich search</a></span><!--  Array
(
    [slug] =&gt; knowledge-rich search
    [parent] =&gt; 
    [name] =&gt; knowledge-rich search
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; A knowledge-rich search uses some form of semantic or structured understanding of the problem space in order to improve the accureacy or speed of {{search}}.  For example, if the problem space is a graph of direct road-travel times between cities, then a knowledege-rich serach might use the crow-flies distance between the cities as a {{heuristic}} to guide the search for a shortest path between two locations.  In contrast a low-knowledge or knowledge-poor approach would just take the distances as path costs without taking into account their geographic nature.
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/aibook/kohonen network">Kohonen networks</a></span><!--  Array
(
    [slug] =&gt; kohonen network
    [parent] =&gt; 
    [name] =&gt; Kohonen networks
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; A Kohonen network is a {{neural network}} algorithm that creates a {{self-organising map}}.  A grid of nodes (usually 2D regular mesh) is initialised with a vector a eaqch node. Then one by one the training data is compared with the nodes and allocated to the node that is closest.  The node itself is swung to be closer to the new example, as are the nodes immediately surrounding it.  Sometimes those a little way off are negatively weighted (pushed away from) the training set item.  Iterating this process leads to a set of node values on the grid such that close cells tend to have some sort of semantic unity in terms of how they classify the data.
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/aibook/labelling">labelling</a></span><!--  Array
(
    [slug] =&gt; labelling
    [parent] =&gt; 
    [name] =&gt; labelling
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; Labelling is the process of giving data items some form of label, often a {{classification}}.  In {{supervised machine learning}}, {{human labelling}} is often necessary for training, but sometimes there may be an automatic labelling, for example, if there are  multiple kinds of sensors, or post-hoc knowledge about outcomes.  The execution phase of {{machine learning}} often involves allocating labels to {{unlabelled data}}.
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/aibook/laird, john">Laird, John</a></span><!--  Array
(
    [slug] =&gt; laird, john
    [parent] =&gt; 
    [name] =&gt; Laird, John
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; John Laird is a Perofessor of Computer Science at Michigan State University, is the creator of the {{SOAR}} {{cognitive architecture}}.
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/aibook/lamda">LaMDA </a></span><!--  Array
(
    [slug] =&gt; lamda
    [parent] =&gt; 
    [name] =&gt; LaMDA 
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; LaMDA is a conversational {{large-language model}} developed by {{Google}}.
)
  --></li>
  <li ><span class="name"><a href="https://alandix.com/glossary/aibook/laplacian operator">Laplacian operator</a></span><!--  Array
(
    [slug] =&gt; laplacian operator
    [parent] =&gt; 
    [name] =&gt; Laplacian operator
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; 
)
  --></li>
  <li ><span class="name"><a href="https://alandix.com/glossary/aibook/laplacian-of-gaussian filter">Laplacian-of-Gaussian filter</a></span><!--  Array
(
    [slug] =&gt; laplacian-of-gaussian filter
    [parent] =&gt; 
    [name] =&gt; Laplacian-of-Gaussian filter
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; 
)
  --></li>
  <li ><span class="name">large language model meta AI</span> - See <span class="has_content"><a href="https://alandix.com/glossary/aibook/llama">LLaMA</a></span><!--  Array
(
    [slug] =&gt; large language model meta ai
    [parent] =&gt; 
    [name] =&gt; large language model meta AI
    [canonical] =&gt; 
    [see] =&gt; llama
    [description] =&gt; 
)
  --></li>
  <li ><span class="name"><a href="https://alandix.com/glossary/aibook/large language model">large language model</a></span><!--  Array
(
    [slug] =&gt; large language model
    [parent] =&gt; 
    [name] =&gt; large language model
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; 
)
  --></li>
  <li ><span class="name"><a href="https://alandix.com/glossary/aibook/latent semantic analysis">latent semantic analysis</a></span><!--  Array
(
    [slug] =&gt; latent semantic analysis
    [parent] =&gt; 
    [name] =&gt; latent semantic analysis
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; 
)
  --></li>
  <li ><span class="name"><a href="https://alandix.com/glossary/aibook/latent space">latent space</a></span><!--  Array
(
    [slug] =&gt; latent space
    [parent] =&gt; 
    [name] =&gt; latent space
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; 
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/aibook/lateral inhibition">lateral inhibition</a></span><!--  Array
(
    [slug] =&gt; lateral inhibition
    [parent] =&gt; 
    [name] =&gt; lateral inhibition
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; Lateral ihibition is when a groups of nodes at the same layer in a brain or {{artficial neural netwrok}} have interconnections that inhibit one another.  This is may encode aspects where there are different interprettions of a visual scene or data that are effectively fighting for each other.  The blob to the right fo the scene is a pussy cat or a seal, but can't be both at the same time.
)
  --></li>
  <li ><span class="name"><a href="https://alandix.com/glossary/aibook/learning rate">learning rate</a></span><!--  Array
(
    [slug] =&gt; learning rate
    [parent] =&gt; 
    [name] =&gt; learning rate
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; 
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/aibook/least-squares">least-squares</a></span><!--  Array
(
    [slug] =&gt; least-squares
    [parent] =&gt; 
    [name] =&gt; least-squares
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; Many {{machine learning}} and {{statistical}} algorithms can be formulated as {{least square}} procedures.  They are effectively minimising the sum of the squares of the differnce between the result they deliver and the desired result -- called the {{mean squared error}}.  In some case this is explicit in the way the algorithm is forumulated, for example the {{backpropagation}} algorithm is a form of {{gradient descent}} that calculates the changes needed to each weight to make the best reduction in the mean square.  Simularly, {{linear rgeression}} minimises the sum of squares of the {{residuals}} (that is the difference between the fitted {[regression line}} and the actual data points).
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/aibook/lee sedol">Lee Sedol</a></span><!--  Array
(
    [slug] =&gt; lee sedol
    [parent] =&gt; 
    [name] =&gt; Lee Sedol
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; Lee Se-dol is the Go grandmaster who was defeated by {{AlphaGo}} in 2016.
)
  --></li>
  <li ><span class="name"><a href="https://alandix.com/glossary/aibook/lego-style matching">lego-style matching</a></span><!--  Array
(
    [slug] =&gt; lego-style matching
    [parent] =&gt; 
    [name] =&gt; lego-style matching
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; 
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/aibook/lemoine, blake">Lemoine, Blake</a></span><!--  Array
(
    [slug] =&gt; lemoine, blake
    [parent] =&gt; 
    [name] =&gt; Lemoine, Blake
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; Blake Lemoine was  a Google engineer who was suspended in 2022 and later fired because he had claimed that the Google chatbot {{LaMDA}} was sentient 
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/aibook/lenat, doug">Lenat, Doug</a></span><!--  Array
(
    [slug] =&gt; lenat, doug
    [parent] =&gt; 
    [name] =&gt; Lenat, Doug
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; Doug Lenart was a visionary AI researcher who founded the long-running {{CYC project}} to codify and reason about general knowledge.
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/aibook/lex">Lex</a></span><!--  Array
(
    [slug] =&gt; lex
    [parent] =&gt; 
    [name] =&gt; Lex
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; Lex is a notation and Unix command-line programme to generate {{lexical analysers}}.  It uses regular expressions to match valid tokens, such as an variable name or number.  It is usualy used as a pre-processor before syntax analysis.
)
  --></li>
  <li ><span class="name"><a href="https://alandix.com/glossary/aibook/lexical processing">lexical processing</a></span><!--  Array
(
    [slug] =&gt; lexical processing
    [parent] =&gt; 
    [name] =&gt; lexical processing
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; 
)
  --></li>
  <li ><span class="name"><a href="https://alandix.com/glossary/aibook/lexicon">lexicon</a></span><!--  Array
(
    [slug] =&gt; lexicon
    [parent] =&gt; 
    [name] =&gt; lexicon
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; 
)
  --></li>
  <li ><span class="name"><a href="https://alandix.com/glossary/aibook/liberal suggestions">liberal suggestions</a></span><!--  Array
(
    [slug] =&gt; liberal suggestions
    [parent] =&gt; 
    [name] =&gt; liberal suggestions
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; 
)
  --></li>
  <li ><span class="name"><a href="https://alandix.com/glossary/aibook/licklider, j.c.r">Licklider, J.C.R.</a></span><!--  Array
(
    [slug] =&gt; licklider, j.c.r
    [parent] =&gt; 
    [name] =&gt; Licklider, J.C.R.
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; 
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/aibook/ligo">LiGO</a></span><!--  Array
(
    [slug] =&gt; ligo
    [parent] =&gt; 
    [name] =&gt; LiGO
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; LiGO {Linear Growth Operator} is a method to reduce the computation needed to train {{large language models} thus both improving computational effeiciency and reducing environmental cost.
)
  --></li>
  <li ><span class="name"><a href="https://alandix.com/glossary/aibook/limb control">limb control</a></span><!--  Array
(
    [slug] =&gt; limb control
    [parent] =&gt; 
    [name] =&gt; limb control
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; 
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/aibook/linear discriminant analysis">linear discriminant analysis</a></span><!--  Array
(
    [slug] =&gt; linear discriminant analysis
    [parent] =&gt; 
    [name] =&gt; linear discriminant analysis
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; Linear discriminant analysis attemots to create one or more {{linear functions}} to separate known classes.  In the case of two classes and 2D data this is simply a line that runs between the classes.  Ideally the classes are completely separated by the linear functions, but in soem case this may be impossible, either becasue of noise blurring the boundaries (in which case the linear discrimant may still be a good classifier), or becasue the classes are not {{linearly separable}}.
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/aibook/linear filter">linear filter</a></span><!--  Array
(
    [slug] =&gt; linear filter
    [parent] =&gt; 
    [name] =&gt; linear filter
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; Linear filters are functions/algorithms that work on numerical data including{{sequential data}}, such as {{time series}}, and {{images}}, but which only include linear combinations of the input.  Examples include compting a rolling seven-day average for daily data, or the {{Guassian filter}} for images. &lt;br&gt; &amp;nbsp; &amp;nbsp; &amp;nbsp; seven_day&lt;sub&gt;t&lt;/sub&gt; = (d&lt;sub&gt;t&lt;/sub&gt; + d&lt;sub&gt;t&amp;minus;1&lt;/sub&gt; + d&lt;sub&gt;t&amp;minus;2&lt;/sub&gt; + d&lt;sub&gt;t&amp;minus;3&lt;/sub&gt; + d&lt;sub&gt;t&amp;minus;4&lt;/sub&gt; + d&lt;sub&gt;t&amp;minus;5&lt;/sub&gt; + d&lt;sub&gt;t&amp;minus;6&lt;/sub&gt;) / 7
)
  --></li>
  <li ><span class="name">linear growth operator</span> - See <span class="has_content"><a href="https://alandix.com/glossary/aibook/ligo">LiGO</a></span><!--  Array
(
    [slug] =&gt; linear growth operator
    [parent] =&gt; 
    [name] =&gt; linear growth operator
    [canonical] =&gt; 
    [see] =&gt; ligo
    [description] =&gt; 
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/aibook/linearly inseparable">linearly inseparable</a></span><!--  Array
(
    [slug] =&gt; linearly inseparable
    [parent] =&gt; 
    [name] =&gt; linearly inseparable
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; Two sets of data are linearly inseparable if there is no straight line (or {{hyperplane}} in higher dimensions) that spearates them. It is the opposite of {{linear separabile}}.
)
  --></li>
  <li ><span class="name">linear patches</span> - See <span class="has_content"><a href="https://alandix.com/glossary/aibook/piecewise linear">piecewise linear</a></span><!--  Array
(
    [slug] =&gt; linear patches
    [parent] =&gt; 
    [name] =&gt; linear patches
    [canonical] =&gt; 
    [see] =&gt; piecewise linear
    [description] =&gt; 
)
  --></li>
  <li ><span class="name"><a href="https://alandix.com/glossary/aibook/linear planning">linear planning</a></span><!--  Array
(
    [slug] =&gt; linear planning
    [parent] =&gt; 
    [name] =&gt; linear planning
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; 
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/aibook/linear regression">linear regression</a></span><!--  Array
(
    [slug] =&gt; linear regression
    [parent] =&gt; 
    [name] =&gt; linear regression
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; Linear regression is a {{statistical technique}} to fit a best line through 2D (x,y) data.   That is, it finds a line that predicts the y value ({{dependent variable}}) using the x value  ({{independent variable}}).  More formally, given data y&lt;sub&gt;i&lt;/sub&gt;, x&lt;sub&gt;i&lt;/sub&gt;, one is trying to find m and c to fit a line of the form y=mx+c.   The best values of m and c can be calculated using:&lt;br&gt; &amp;nbsp; &amp;nbsp; &amp;nbsp; m &amp;nbsp; = &amp;nbsp; &amp;sigma;&lt;sub&gt;x&lt;/sub&gt;&lt;sub&gt;y&lt;/sub&gt; / &amp;sigma;&lt;sup&gt;2&lt;/sup&gt;&lt;sub&gt;x&lt;/sub&gt;&lt;br&gt; &amp;nbsp; &amp;nbsp; &amp;nbsp; c &amp;nbsp; = &amp;nbsp; &amp;mu;&lt;sub&gt;y&lt;/sub&gt; &amp;minus; m &amp;mu;&lt;sub&gt;x&lt;/sub&gt;  &lt;br&gt; where &amp;mu;&lt;sub&gt;x&lt;/sub&gt; and &amp;mu;&lt;sub&gt;x&lt;/sub&gt; are the {{arithmetic mean}} of x and y respectively, &amp;sigma;&lt;sup&gt;2&lt;/sup&gt;&lt;sub&gt;x&lt;/sub&gt; is the {{variance}} of x, and  &amp;sigma;&lt;sub&gt;x&lt;/sub&gt;&lt;sub&gt;y&lt;/sub&gt; is the covariance of x and y&lt;br&gt; Note that the idea of 'best' here means minimising the sum of the squares of the y error, that is minimising: &lt;br&gt; &amp;nbsp; &amp;nbsp; &amp;nbsp; &amp;Sigma; (y&lt;sub&gt;i&lt;/sub&gt; &amp;minus; mx&lt;sub&gt;i&lt;/sub&gt;+c)&lt;sup&gt;2&lt;/sup&gt; &lt;br&gt; You get different answer if you swop the variables and look for the regression of x on y.  &lt;br&gt;&lt;br&gt; When there is more than one {{independent variable}}, there is an extension, {{multi-linear regression}}, which instead fits a {{hyperplane}}) to the data.
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/aibook/linearly separable">linearly separable</a></span><!--  Array
(
    [slug] =&gt; linearly separable
    [parent] =&gt; 
    [name] =&gt; linearly separable
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; Two sets of data are linearly separable if there is a straight line (or {{hyperplane}} in higher dimensions) that separates them.  In other words there is a linear combination (a&lt;sub&gt;i&lt;/sub&gt;) of the features and a {{threshold}} T such that fp&lt;br&gt; &amp;nbsp;   &amp;nbsp;   &amp;nbsp;   a&lt;sub&gt;1&lt;/sub&gt;d&lt;sub&gt;1&lt;/sub&gt; +  a&lt;sub&gt;2&lt;/sub&gt;d&lt;sub&gt;2&lt;/sub&gt; + ...  + a&lt;sub&gt;n&lt;/sub&gt;d&lt;sub&gt;n&lt;/sub&gt;  is less than T for data items (d&lt;sub&gt;i&lt;/sub&gt;) in the first datasets and greater than T for the second &lt;br&gt; In such cases, the {{hyperplane}} acts as a {[classifier}}.
)
  --></li>
  <li ><span class="name"><a href="https://alandix.com/glossary/aibook/link structure">link structure</a></span><!--  Array
(
    [slug] =&gt; link structure
    [parent] =&gt; 
    [name] =&gt; link structure
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; 
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/aibook/lisp">Lisp</a></span><!--  Array
(
    [slug] =&gt; lisp
    [parent] =&gt; 
    [name] =&gt; Lisp
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; Lisp is one of the most significant early AI programming langauges, known for its heavy use of brackets.  Lisp stands for &quot;List processing&quot; and the language is heavily influenced by the {{lambda calculus}}.  Lisp has several dialects including {{Scheme}}.  Lisp is an early {{functional programmon g}} lanugage although not a {{pure finctional language}} as it has updatabale variables and state.  
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/aibook/llama">LLaMA</a></span><!--  Array
(
    [slug] =&gt; llama
    [parent] =&gt; 
    [name] =&gt; LLaMA
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; LLaMA (large language model meta AI) is a {{foundation model}} released by {{Meta}}.  It was designed to be cheaper to initially train and fine tune than other {{large language models}} at the time.  The second iteration, LLaMA 2 is open source and this has allowed substantial further innovation.
)
  --></li>
  <li class="has_content"><span class="name">loans example</span> - See <span class="has_content"><a href="https://alandix.com/glossary/aibook/production system">production system</a></span><!--  Array
(
    [slug] =&gt; loans example
    [parent] =&gt; 
    [name] =&gt; loans example
    [canonical] =&gt; 
    [see] =&gt; production system
    [description] =&gt; This is an example {{production system}} about loan approvals used in  the AI textbook.  The example is coded in {{Prolog}} in the code examples for the book.
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/aibook/local data access">local data access</a></span><!--  Array
(
    [slug] =&gt; local data access
    [parent] =&gt; 
    [name] =&gt; local data access
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; Local data access is when the data required for code to execute is physiacally close to it, in the sense of on the same proessor, or at least close network.  This is an issue for {{big data}} processing especially on {{cloud computing}} platforms, where data may be stored in multiple computers often in physically remote {{data centres}}.  The goal is to perfom as much procesisng locally without moving data around.   Where the code is accessing ultiple data items, it is important that these are then stored physically close to one another.  This may be because theer are interincis commonlaity about data so they can be permamnently stored close to one another, or brought together for the purposes of processing. An exmaple all the former is data about a single person, and maybe summary inorrmation about their friends, in a {{social network}} which can be stored as a single record or data item.  An example of the latter is {{MapReduce}}, which uses a common {{hash}} value to bring items that need to be processed together by the {{reduce}} step on the same processor.
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/aibook/local maximum">local maximum</a></span><!--  Array
(
    [slug] =&gt; local maximum
    [parent] =&gt; 
    [name] =&gt; local maximum
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; A local maximum for a function is a set of parameters for the function that is the largest in a neighbourhood, that is f(p) &amp;gt; f(p+&amp;delta;) for small changes $delta; in the parametre value.  In a physical landscape this would correspond to the top of a small hill, but where there may be a much larger hill elsewhere.   This geographic analogy can be helpful in the case of the {{firness landcape}} egnerated by a {{fitness function}} for {{machine learning}}; however, whlst the space of parametres for a physicla lcation is 2D (latitued,longitude), the space of parameters of a {{ftness functoon}} is often {{multi-dimensional}} and may include a mix of different kinds of parameters, not just numeric.   It is the opposite of a {{local minimum}} and is to be contrasted with a {{global maximum}} which is the largest/highest point overall.
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/aibook/local minima/maxima">local minima/maxima</a></span><!--  Array
(
    [slug] =&gt; local minima/maxima
    [parent] =&gt; 
    [name] =&gt; local minima/maxima
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; In a {[fitness landscape}} there may be many {{local minima}} and {{local maxima}} that is points the locally are smaller or larger than neoghbouring points, but that are not the smallest or bggest overall.
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/aibook/local minimum">local minimum</a></span><!--  Array
(
    [slug] =&gt; local minimum
    [parent] =&gt; 
    [name] =&gt; local minimum
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; A local minimum for a function is a set of parameters for the function that is the smallest in a neighbourhood, that is f(p) &amp;lt; f(p+&amp;delta;) for small changes $delta; in the parameter value.  In a physical landscape this would correspond to the bottom of a small hole, but where there may be a much deeper gulf.   It is the opposite of a {{local maximum}} and is to be contrasted with a {{global mimumum}} which is the smallest/lowest point overall.
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/aibook/local planning">local planning</a></span><!--  Array
(
    [slug] =&gt; local planning
    [parent] =&gt; 
    [name] =&gt; local planning
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; Local planning is about dealing with small scale navigation including {{obstacle avoidance}} or navigating through doorways or corridors.  It s in contrast with {{global planning}} which is about finding the best overall route, but ignores small details. 
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/aibook/locality">locality</a></span><!--  Array
(
    [slug] =&gt; locality
    [parent] =&gt; 
    [name] =&gt; locality
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; We say an algorithm ir problem to be solved has locality, whe the date requred for each step is close to one another in memory.  This depends on the way in whch data is stored in memory, so that a good choice of {{knowledge representation}} can make algoritms ore effeicient.  However, some problems, particualy those involving complex {{graph processong}}, are intrinsically non-local -- no representation can entirely avod dusant memory accesses. This is particularly important when dealing with {{big data}} distributed over many storage devices or even, in {{cloud computing}} multiple {{data centres}} as non-local data access may involve substantial network traffic and delays waiting for it.  However, it can also be important in {{parallel processing}} on {{GPU}} chips, or improving cache performance  on single computers. 
)
  --></li>
  <li ><span class="name"><a href="https://alandix.com/glossary/aibook/logarithm base">logarithm base</a></span><!--  Array
(
    [slug] =&gt; logarithm base
    [parent] =&gt; 
    [name] =&gt; logarithm base
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; 
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/aibook/logarithmic transform">logarithmic transform</a></span><!--  Array
(
    [slug] =&gt; logarithmic transform
    [parent] =&gt; 
    [name] =&gt; logarithmic transform
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; Taking the logorithm of data items can sometimes be a useful form of data transformation in several circumstances.  A logorithmic transform turns exponential growth onto linear grpwth, so can be useful for long term financial data where value of money gradually reduces in value due to inflation.  It can also be used where the variability grows with the size of data items, which is freqently the case for certain kinds of time measurements.  If you are measuring time delays often larger values also have larger variability.  If the logarithm of data follows the {{Normal distribution}}, the data is known as {{log-normal}}.
)
  --></li>
  <li ><span class="name"><a href="https://alandix.com/glossary/aibook/logic">logic</a></span>- has subentries<ul class="subentries"><li ><span class="name"><a href="https://alandix.com/glossary/aibook/logic!as knowledge representation">as knowledge representation</a></span><!--  Array
(
    [slug] =&gt; logic!as knowledge representation
    [parent] =&gt; logic
    [name] =&gt; as knowledge representation
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; 
)
  --></li>
<li ><span class="name">fuzzy</span> - See <span ><a href="https://alandix.com/glossary/aibook/fuzzy logic">fuzzy logic</a></span><!--  Array
(
    [slug] =&gt; logic!fuzzy
    [parent] =&gt; logic
    [name] =&gt; fuzzy
    [canonical] =&gt; 
    [see] =&gt; fuzzy logic
    [description] =&gt; 
)
  --></li>
<li ><span class="name"><a href="https://alandix.com/glossary/aibook/logic!operators">operators</a></span><!--  Array
(
    [slug] =&gt; logic!operators
    [parent] =&gt; logic
    [name] =&gt; operators
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; 
)
  --></li>
<li ><span class="name">standard</span> - See <span class="has_content"><a href="https://alandix.com/glossary/aibook/standard logic">standard logic</a></span><!--  Array
(
    [slug] =&gt; logic!standard
    [parent] =&gt; logic
    [name] =&gt; standard
    [canonical] =&gt; 
    [see] =&gt; standard logic
    [description] =&gt; 
)
  --></li></ul><!--  Array
(
    [slug] =&gt; logic
    [parent] =&gt; 
    [name] =&gt; logic
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; 
    [subentries] =&gt; Array
        (
            [logic!as knowledge representation] =&gt; Array
                (
                    [slug] =&gt; logic!as knowledge representation
                    [parent] =&gt; logic
                    [name] =&gt; as knowledge representation
                    [canonical] =&gt; 
                    [see] =&gt; 
                    [description] =&gt; 
                )

            [logic!fuzzy] =&gt; Array
                (
                    [slug] =&gt; logic!fuzzy
                    [parent] =&gt; logic
                    [name] =&gt; fuzzy
                    [canonical] =&gt; 
                    [see] =&gt; fuzzy logic
                    [description] =&gt; 
                )

            [logic!operators] =&gt; Array
                (
                    [slug] =&gt; logic!operators
                    [parent] =&gt; logic
                    [name] =&gt; operators
                    [canonical] =&gt; 
                    [see] =&gt; 
                    [description] =&gt; 
                )

            [logic!standard] =&gt; Array
                (
                    [slug] =&gt; logic!standard
                    [parent] =&gt; logic
                    [name] =&gt; standard
                    [canonical] =&gt; 
                    [see] =&gt; standard logic
                    [description] =&gt; 
                )

        )

)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/aibook/logistic function">logistic function</a></span><!--  Array
(
    [slug] =&gt; logistic function
    [parent] =&gt; 
    [name] =&gt; logistic function
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; A logistic function is a form of {{sigmoid functiion}} often used as a weak {[threshold}} in an {{neural network}}.  Like a {{step function}} it maps unbounded values to a finite [0,1] range and is zero for large negatuve values and one for lareg postuve values.  However, unlike a  {{step function}} the prcess is smooth with a near {{linear}} section in the middle and gently asymptotes for more edxtreme values.  This is imprtant for {{machine learning}} as, in general, it is easier to learn {[contnuous}} features. 
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/aibook/long-tail distribution">long-tail distribution</a></span><!--  Array
(
    [slug] =&gt; long-tail distribution
    [parent] =&gt; 
    [name] =&gt; long-tail distribution
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; A long-tail distribution is a form of {{probability distribution}} or empirical histogram of values, where the number of extremely large elements may be sufficient to make a major difference to common measures such as the {{mean}} or {[standard deviation}}.  A common example of a long-tailed distribution is the {{power-law distrubution}} often encounterd in {{network data}} such as  {{social media}} connections or {{web links}}.  Income distrubutions in most soieties are less extreme than the power law, and usually has a well-defined mean but still have effectively unbounded {{variance}}.  Note that some distributions such as the {{Normal distribution}} may have a long tail in the sense that there is the potential for a few very large values, but these are exteremely rare.
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/aibook/long-term memory">long-term memory</a></span><!--  Array
(
    [slug] =&gt; long-term memory
    [parent] =&gt; 
    [name] =&gt; long-term memory
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; Long-term memory in humans refers to our memory of events, people, ideas and skills that stay with us for very long periods, perhaps for our whole lives.  It is often divided into: {{episodic memory}} of events and gtheor relations; {[declarative memory}} of facts and concepts; and {{procedural memory}} of how to do things.  In addition, {{decalartuve memory}} for different kinds of things, for example perosnal relationships vs public figures, are stored in different ways.
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/aibook/long-term potentiation">long-term potentiation</a></span><!--  Array
(
    [slug] =&gt; long-term potentiation
    [parent] =&gt; 
    [name] =&gt; long-term potentiation
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; Long-term potentiation refers to the way in which chemcial changes in individual {{neurones}} can alter firing (either making it easier or harder) for a period of time from seconds to hours.  This can give rise to {{mezzanine memory}} or {{long-term working memory}} that lasts over periods of up to a few hours.  In contrast, human {{short term memory}} and {[conscious reasoning}} are deternines by very short term electrical (ionic activity) and {{long-term memory}} is based on actual growth or decay of the {{axions}} and {{synapses}} that connect {{neurones}}.
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/aibook/long-term short-term memory networks">long-term short-term memory networks</a></span><!--  Array
(
    [slug] =&gt; long-term short-term memory networks
    [parent] =&gt; 
    [name] =&gt; long-term short-term memory networks
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; Long-term short-term memory networks (LTSM) are used on {{time series data}}.  LTSM have special memory nodes that can be triggered to remember the state of their input based on the values of other parts of a {{neural network}}.  This can be used to model effects such as systems that have long phases of similar behaviour, with occasional changes to other kinds of behaviour, or to recall important events for long periods.
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/aibook/lora">LoRA</a></span><!--  Array
(
    [slug] =&gt; lora
    [parent] =&gt; 
    [name] =&gt; LoRA
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; LoRA (low-rank adaptation of large language models} performs {{dimensional reduction}} on inner layers of the {{deep neural network}} used in a {{large language model}} thus reducing the cost of execution and retraining.  This has economic and environmental advantages.
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/aibook/lovelace, ada">Lovelace, Ada</a></span><!--  Array
(
    [slug] =&gt; lovelace, ada
    [parent] =&gt; 
    [name] =&gt; Lovelace, Ada
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; Ada Lovelace was a collaborator with {{Charles Babbage}} in the 18th Century, and daughter of the poet Lord Byron.  She is regarded as the first programmer, designing algorithsm that would run on the {{Analytical Engine}},. 
 Critically also she was an early philosopher of computation, identifiying the mportance of &lt;em&gt;symbolic results&lt;/em&gt; and difference between calcuation and computation:&lt;br&gt; 
&quot;&lt;em&gt;The {{Difference Engine}} can merely tabulate, and is incapable of developing, the {{Analytical Engine}} can either tabulate or develope&lt;/em&gt;&quot;
&lt;br&gt;
She also wrpte in aesthetic terms appreciating the  &quot;&lt;em&gt;intrinsic beauty, symmetry and logical completeness&lt;\em&gt;&quot; of mathemtatcal sciences  and the way that the {{Analytical Engine}} &quot;&lt;em&gt;weaves algebraical patterns just as the {{Jacquard-loom}} weaves flowers and leaves&lt;/em&gt;&quot;.
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/aibook/lower bound">lower bound</a></span><!--  Array
(
    [slug] =&gt; lower bound
    [parent] =&gt; 
    [name] =&gt; lower bound
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; A lower bound is an estimate, {{heuristic}} or calculated value that yoo know lies at or below the smallest possible value of an unknown value or function. For example, if you are looking for the road distance between two locations, you know this will be at least as large as the {{Euclidean dstace}} between them (crow flies distance).  Knowing a lower bound can sometimes allow whole branches of a {{search tree}} or similar structure to be {{pruned}}.  For example, if you are looking for the smallest value in a tree and have found a node with value 7, and then consider an alternative branch in the tree where a {{heuristic}} tells you that the {{lower bound}} for nodes in the branch is 12, you can safely ignore the entre sub-tree.
)
  --></li>
  <li ><span class="name"><a href="https://alandix.com/glossary/aibook/machine learning">machine learning</a></span>- has subentries<ul class="subentries"><li ><span class="name"><a href="https://alandix.com/glossary/aibook/machine learning!application">application</a></span><!--  Array
(
    [slug] =&gt; machine learning!application
    [parent] =&gt; machine learning
    [name] =&gt; application
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; 
)
  --></li>
<li ><span class="name"><a href="https://alandix.com/glossary/aibook/machine learning!training">training</a></span><!--  Array
(
    [slug] =&gt; machine learning!training
    [parent] =&gt; machine learning
    [name] =&gt; training
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; 
)
  --></li>
<li ><span class="name"><a href="https://alandix.com/glossary/aibook/machine learning!validation">validation</a></span><!--  Array
(
    [slug] =&gt; machine learning!validation
    [parent] =&gt; machine learning
    [name] =&gt; validation
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; 
)
  --></li></ul><!--  Array
(
    [slug] =&gt; machine learning
    [parent] =&gt; 
    [name] =&gt; machine learning
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; 
    [subentries] =&gt; Array
        (
            [machine learning!application] =&gt; Array
                (
                    [slug] =&gt; machine learning!application
                    [parent] =&gt; machine learning
                    [name] =&gt; application
                    [canonical] =&gt; 
                    [see] =&gt; 
                    [description] =&gt; 
                )

            [machine learning!training] =&gt; Array
                (
                    [slug] =&gt; machine learning!training
                    [parent] =&gt; machine learning
                    [name] =&gt; training
                    [canonical] =&gt; 
                    [see] =&gt; 
                    [description] =&gt; 
                )

            [machine learning!validation] =&gt; Array
                (
                    [slug] =&gt; machine learning!validation
                    [parent] =&gt; machine learning
                    [name] =&gt; validation
                    [canonical] =&gt; 
                    [see] =&gt; 
                    [description] =&gt; 
                )

        )

)
  --></li>
  <li ><span class="name"><a href="https://alandix.com/glossary/aibook/macro languages">macro languages</a></span><!--  Array
(
    [slug] =&gt; macro languages
    [parent] =&gt; 
    [name] =&gt; macro languages
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; 
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/aibook/magic square">magic square</a></span><!--  Array
(
    [slug] =&gt; magic square
    [parent] =&gt; 
    [name] =&gt; magic square
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; A magic square is a square grid of numbers, typically consecutuve, where each row, column and sometimes diagnonal add up to the same number.
)
  --></li>
  <li ><span class="name">magic squares</span> - See <span class="has_content"><a href="https://alandix.com/glossary/aibook/magic square">magic square</a></span><!--  Array
(
    [slug] =&gt; magic squares
    [parent] =&gt; 
    [name] =&gt; magic squares
    [canonical] =&gt; 
    [see] =&gt; magic square
    [description] =&gt; 
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/aibook/manhattan block distance">Manhatten block distance</a></span><!--  Array
(
    [slug] =&gt; manhattan block distance
    [parent] =&gt; 
    [name] =&gt; Manhatten block distance
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; The Manhatten block distance simply adds the x and y distance like walking the streets in a city.  In higher dimensions it is simply the sum of the absolute distance along every dimension, or in data terms the sum of the difference of every {{feature}}.  Alternative distance measures include the {{Euclidean distance}}, which is the 'as the crows flies' distance between two point
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/aibook/map">map</a></span><!--  Array
(
    [slug] =&gt; map
    [parent] =&gt; 
    [name] =&gt; map
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; Map is a standard function in {{Lisp}} and other {{functional programming}} languages.  Map has two arguments map(f,list). where the function f is applied to every element of the list forming a new list.  That is &lt;br&gt; map( f, (a,b,c...z) )  =  ( f(a), f(b), f(c), ... f(z) ).  &lt;br&gt;&lt;br&gt; The map function is one of the inspirations for the {{MapReduce}} infrastructure for massive scale processing of {{big data}}.
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/aibook/mapreduce">MapReduce</a></span><!--  Array
(
    [slug] =&gt; mapreduce
    [parent] =&gt; 
    [name] =&gt; MapReduce
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; MapReduce is a distributed computing framework for delaing with {{big data}}.  It is inspired by the {{map}} and {{reduce}} functions found in {{Lisp}} and other {{functional programming}} languages: the {{map}} stage initially processes each item of data separately and then the {[reduce}} stage collects and aggregates related intermediate results.  However, MapReduce has two crucial  additional features : (i) the use of a hash as means to distribute intermediate results over different computers, so redcuing the likelihood of {{Byzantine conditions}}; and (ii) means to monitor for failure of individual computers and recvover from this.    &lt;br&gt;&lt;table align=&quot;center&quot;&gt;   &lt;tr&gt;&lt;td&gt;each data item&lt;/td&gt;&lt;td&gt;&amp;rarr; &lt;/td&gt;&lt;td align=&quot;center&quot;&gt;map &lt;/td&gt;&lt;td&gt;&amp;rarr; &lt;/td&gt; &lt;td&gt;hash + processed data&lt;/td&gt;&lt;/tr&gt;    &lt;tr&gt;&lt;td&gt;all data for a hash&lt;/td&gt;&lt;td&gt;&amp;rarr; &lt;/td&gt;&lt;td align=&quot;center&quot;&gt;reduce &lt;/td&gt;&lt;td&gt;&amp;rarr; &lt;/td&gt; &lt;td&gt;one or more agggregated calculations for the hashed data &lt;br&gt;(possibly including fresh has for further reduce steps)&lt;/td&gt;&lt;/tr&gt;    &lt;/table&gt;.    &lt;br&gt;MapReduce was initially developed by engineers at Google, but has since become part of the open source Apache {{Hadoop}} project.  
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/aibook/markov, andrey">Markov, Andrey</a></span><!--  Array
(
    [slug] =&gt; markov, andrey
    [parent] =&gt; 
    [name] =&gt; Markov, Andrey
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; Andrey Andreyevich Markov was a Russian mathematician, especially well known for the concept of a {{Markov model}} or {{Markov process}}
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/aibook/markov model">Markov model</a></span><!--  Array
(
    [slug] =&gt; markov model
    [parent] =&gt; 
    [name] =&gt; Markov model
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; A Markov model is a {{stochastic process}} where the probability of each step in the sequence is dependent only on the previous step, or in the case of {{higher-order Markov models}} the last n-steps.  Markov models are frequently shown as matrices where each entry in the matrix M&lt;sub&gt;ik&lt;/sub&gt; represents the {{conditional probability}} that if the current state is i, then the next state will be k.
)
  --></li>
  <li ><span class="name"><a href="https://alandix.com/glossary/aibook/marr's primal sketch">Marr's primal sketch</a></span><!--  Array
(
    [slug] =&gt; marr's primal sketch
    [parent] =&gt; 
    [name] =&gt; Marr's primal sketch
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; 
)
  --></li>
  <li ><span class="name">Mary Shelley</span> - See <span class="has_content"><a href="https://alandix.com/glossary/aibook/shelley, mary">Shelley, Mary</a></span><!--  Array
(
    [slug] =&gt; mary shelley
    [parent] =&gt; 
    [name] =&gt; Mary Shelley
    [canonical] =&gt; 
    [see] =&gt; shelley, mary
    [description] =&gt; 
)
  --></li>
  <li ><span class="name">materialism</span> - See <span class="has_content"><a href="https://alandix.com/glossary/aibook/physicalism">physicalism</a></span><!--  Array
(
    [slug] =&gt; materialism
    [parent] =&gt; 
    [name] =&gt; materialism
    [canonical] =&gt; 
    [see] =&gt; physicalism
    [description] =&gt; 
)
  --></li>
  <li ><span class="name">MDS</span> - See <span class="has_content"><a href="https://alandix.com/glossary/aibook/multi-dimensional scaling">multi-dimensional scaling</a></span><!--  Array
(
    [slug] =&gt; mds
    [parent] =&gt; 
    [name] =&gt; MDS
    [canonical] =&gt; 
    [see] =&gt; multi-dimensional scaling
    [description] =&gt; 
)
  --></li>
  <li ><span class="name"><a href="https://alandix.com/glossary/aibook/means-ends analysis">means-ends analysis</a></span><!--  Array
(
    [slug] =&gt; means-ends analysis
    [parent] =&gt; 
    [name] =&gt; means-ends analysis
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; 
)
  --></li>
  <li ><span class="name"><a href="https://alandix.com/glossary/aibook/memorizing">memorizing</a></span><!--  Array
(
    [slug] =&gt; memorizing
    [parent] =&gt; 
    [name] =&gt; memorizing
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; 
)
  --></li>
  <li ><span class="name"><a href="https://alandix.com/glossary/aibook/memory">memory</a></span><!--  Array
(
    [slug] =&gt; memory
    [parent] =&gt; 
    [name] =&gt; memory
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; 
)
  --></li>
  <li ><span class="name"><a href="https://alandix.com/glossary/aibook/memristor">memristor</a></span><!--  Array
(
    [slug] =&gt; memristor
    [parent] =&gt; 
    [name] =&gt; memristor
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; 
)
  --></li>
  <li ><span class="name"><a href="https://alandix.com/glossary/aibook/method acting">method acting</a></span><!--  Array
(
    [slug] =&gt; method acting
    [parent] =&gt; 
    [name] =&gt; method acting
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; 
)
  --></li>
  <li ><span class="name"><a href="https://alandix.com/glossary/aibook/microdata">microdata</a></span><!--  Array
(
    [slug] =&gt; microdata
    [parent] =&gt; 
    [name] =&gt; microdata
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; 
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/aibook/microformats">microformats</a></span><!--  Array
(
    [slug] =&gt; microformats
    [parent] =&gt; 
    [name] =&gt; microformats
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; Microformats are ways to encode {{semantic knowledge}} into web pages.  Special classes are used around elements, such as dates, personal details, etc.  For example: &lt;br&gt; &amp;lt;span class=&quot;p-name&quot;&amp;gt;Alan Dix&amp;lt/span&amp;gt &lt;br&gt; This means that a computer reading the page can interpret the marked elements, hence turning the human-readable web page into machine-readable data.
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/aibook/microtubules">microtubules</a></span><!--  Array
(
    [slug] =&gt; microtubules
    [parent] =&gt; 
    [name] =&gt; microtubules
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; Microtubules are tiny structures within biological cells that act rather like scaffolding.  They are what enable single celled organisms, such as amoeba, to move and also pull chromosomes apart and re-assemble them during cell division.  {{Roger Penrose}} has suggested that {{quantum effects}} within microtubules may be related to {{human intelligence}}.
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/aibook/minimax search">minimax search</a></span><!--  Array
(
    [slug] =&gt; minimax search
    [parent] =&gt; 
    [name] =&gt; minimax search
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; Minimax is a search algorithm for playing {{zero-sum games}}.  At each stage, the player whose turn it is looks ahead at each possible move and works out the maximum score the opponent could get if it plays optimally.  The player then chooses the move that minimises the opponents score and hence maxmises their own.  The same algorithm is applied for working out the opponent's move down each branch and so on.
)
  --></li>
  <li ><span class="name"><a href="https://alandix.com/glossary/aibook/misinformation">misinformation</a></span><!--  Array
(
    [slug] =&gt; misinformation
    [parent] =&gt; 
    [name] =&gt; misinformation
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; 
)
  --></li>
  <li ><span class="name"><a href="https://alandix.com/glossary/aibook/modal">modal</a></span><!--  Array
(
    [slug] =&gt; modal
    [parent] =&gt; 
    [name] =&gt; modal
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; 
)
  --></li>
  <li ><span class="name"><a href="https://alandix.com/glossary/aibook/mode">mode</a></span><!--  Array
(
    [slug] =&gt; mode
    [parent] =&gt; 
    [name] =&gt; mode
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; 
)
  --></li>
  <li ><span class="name"><a href="https://alandix.com/glossary/aibook/monotonic reasoning">monotonic reasoning</a></span><!--  Array
(
    [slug] =&gt; monotonic reasoning
    [parent] =&gt; 
    [name] =&gt; monotonic reasoning
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; 
)
  --></li>
  <li ><span class="name">monte carlo algorithm</span> - See <span class="has_content"><a href="https://alandix.com/glossary/aibook/monte carlo techniques">Monte Carlo techniques</a></span><!--  Array
(
    [slug] =&gt; monte carlo algorithm
    [parent] =&gt; 
    [name] =&gt; monte carlo algorithm
    [canonical] =&gt; 
    [see] =&gt; monte carlo techniques
    [description] =&gt; 
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/aibook/monte carlo search">Monte Carlo search</a></span><!--  Array
(
    [slug] =&gt; monte carlo search
    [parent] =&gt; 
    [name] =&gt; Monte Carlo search
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; A Monte Carlo search uses random choices as part of the search algorithm.  An extreme case is random {{generate and test}}, choosing points in the {{search space}} at random until one satisfies the search criterion.   However, forms of Monte Carlo search can also use {{heuristics}} to nudge the algorothm towards more likely parts of the {{search space}}.
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/aibook/monte carlo techniques">Monte Carlo techniques</a></span><!--  Array
(
    [slug] =&gt; monte carlo techniques
    [parent] =&gt; 
    [name] =&gt; Monte Carlo techniques
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; Monte Carlo techniques use some form of random choices as part of their algorithms.  Examples include {{random forests}} and {{Monte Carlo search}}.  These can be practical algorithms, especially where the choice is not entirely random, but guided by {{heuristics}} to enable different chocies to have different probabilities.   Monto Carlo algroithms can also be used as a base case to compare with other algorithms.  For example, a simple Monte Carlo {{optimisation}} algorithm is to simply choose parameters at random and keep track of the best seen so far; if other algorithms do not perform significantly better than this on a data set, then they are clearly not well suited to the kind of data.
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/aibook/monte carlo tree search">Monte Carlo tree search</a></span><!--  Array
(
    [slug] =&gt; monte carlo tree search
    [parent] =&gt; 
    [name] =&gt; Monte Carlo tree search
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; Monte Carlo tree search is any form of {{Monte Carlo search}} applied to trees.  For example a simple variant of depth search could randomly choose branches to follow, stopping when the path gets to a leaf, or, in the case of very large trees, at a random point before (getting bored!).  As with any {{Monte Carlo algorithm}}, heuristics can be used to assign probabilities to different branches. {{AlphaGo}} used {{Monte Carlo  tree search}} alongside {{heuristics}} generated by {{deep neuaral networks}}.
)
  --></li>
  <li ><span class="name"><a href="https://alandix.com/glossary/aibook/moving average">moving average</a></span><!--  Array
(
    [slug] =&gt; moving average
    [parent] =&gt; 
    [name] =&gt; moving average
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; 
)
  --></li>
  <li ><span class="name"><a href="https://alandix.com/glossary/aibook/moving average models">moving average models</a></span><!--  Array
(
    [slug] =&gt; moving average models
    [parent] =&gt; 
    [name] =&gt; moving average models
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; 
)
  --></li>
  <li ><span class="name"><a href="https://alandix.com/glossary/aibook/moving images">moving images</a></span><!--  Array
(
    [slug] =&gt; moving images
    [parent] =&gt; 
    [name] =&gt; moving images
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; 
)
  --></li>
  <li ><span class="name">moving pictures</span> - See <span ><a href="https://alandix.com/glossary/aibook/moving images">moving images</a></span><!--  Array
(
    [slug] =&gt; moving pictures
    [parent] =&gt; 
    [name] =&gt; moving pictures
    [canonical] =&gt; 
    [see] =&gt; moving images
    [description] =&gt; 
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/aibook/multi-dimensional scaling">multi-dimensional scaling</a></span><!--  Array
(
    [slug] =&gt; multi-dimensional scaling
    [parent] =&gt; 
    [name] =&gt; multi-dimensional scaling
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; Multi-dimensional scaling (MDS) creates a two (usually) dimensioanl represention of data that preserves closeness.  Its input is a {{similarity matrix}} (S&lt;sub&gt;ik&lt;/sub&gt;) or a  {{dissimilarity matrix}} (D&lt;sub&gt;ik&lt;/sub&gt;)  between items an then allocates for each item i coordinates (x&lt;sub&gt;i&lt;/sub&gt;,y&lt;sub&gt;i&lt;/sub&gt;) such that, on average, when the {{Euclidean distance}} between (x&lt;sub&gt;i&lt;/sub&gt;,y&lt;sub&gt;i&lt;/sub&gt;)  and (x&lt;sub&gt;j&lt;/sub&gt;,y&lt;sub&gt;j&lt;/sub&gt;) is small then S&lt;sub&gt;ik&lt;/sub&gt;) is large (or D&lt;sub&gt;ik&lt;/sub&gt;) is small) and vice versa.
)
  --></li>
  <li ><span class="name"><a href="https://alandix.com/glossary/aibook/multi-layer perceptron">multi-layer perceptron</a></span><!--  Array
(
    [slug] =&gt; multi-layer perceptron
    [parent] =&gt; 
    [name] =&gt; multi-layer perceptron
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; 
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/aibook/multi-linear regression">multi-linear regression</a></span><!--  Array
(
    [slug] =&gt; multi-linear regression
    [parent] =&gt; 
    [name] =&gt; multi-linear regression
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; Multi-objective optimisation is used when there are sevral criteri, all of which one would wich to optimise.  For example, a company may want to optimise profit, welfare for its workers and environmental impact.  Typically there is not a way to simultaneously optimise all the criteria so there has to be a trade-off between them. Multi-objective optimisation is hence far more difficult than simple single-criterion {{optimisation}}. One solution is to weight the criteria and optmise the weighted sum.  Other approaches attempt to find some or all {{pareto-optimal}} solutions; that is solutions from which there is no way to improve all citeria.  
)
  --></li>
  <li ><span class="name"><a href="https://alandix.com/glossary/aibook/multi-objective optimisation">multi-objective optimisation</a></span><!--  Array
(
    [slug] =&gt; multi-objective optimisation
    [parent] =&gt; 
    [name] =&gt; multi-objective optimisation
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; 
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/aibook/multimodel interfaces">multimodel interfaces</a></span><!--  Array
(
    [slug] =&gt; multimodel interfaces
    [parent] =&gt; 
    [name] =&gt; multimodel interfaces
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; Multimodel interfaces use several human modes of interaction simultaneously ; for example speaking whilst pointing at a screen and saying &quot;put that there&quot;.
)
  --></li>
  <li ><span class="name"><a href="https://alandix.com/glossary/aibook/multiple images">multiple images</a></span><!--  Array
(
    [slug] =&gt; multiple images
    [parent] =&gt; 
    [name] =&gt; multiple images
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; 
)
  --></li>
  <li ><span class="name">mutual information</span> - See <span class="has_content"><a href="https://alandix.com/glossary/aibook/information gain">information gain</a></span><!--  Array
(
    [slug] =&gt; mutual information
    [parent] =&gt; 
    [name] =&gt; mutual information
    [canonical] =&gt; 
    [see] =&gt; information gain
    [description] =&gt; 
)
  --></li>
  <li ><span class="name"><a href="https://alandix.com/glossary/aibook/mutual inhibition">mutual inhibition</a></span><!--  Array
(
    [slug] =&gt; mutual inhibition
    [parent] =&gt; 
    [name] =&gt; mutual inhibition
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; 
)
  --></li>
  <li ><span class="name"><a href="https://alandix.com/glossary/aibook/mycin">MYCIN</a></span><!--  Array
(
    [slug] =&gt; mycin
    [parent] =&gt; 
    [name] =&gt; MYCIN
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; 
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/aibook/n-gram">n-gram</a></span><!--  Array
(
    [slug] =&gt; n-gram
    [parent] =&gt; 
    [name] =&gt; n-gram
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; In a text or sequence an n-gram is a sequence of n consecutive tokens.   In the case of text treated as a sequence words, a 1-gram is a single word in the text and a 2-gram is a word pair. For example, in this descriotion, &quot;text or sequence a&quot; is a 4-gram.  Typically you are interested in the set of all n-grams in a text  for some n and their respective frequencies of occurrence in the text.
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/aibook/named entity recognition">named entity recognition</a></span><!--  Array
(
    [slug] =&gt; named entity recognition
    [parent] =&gt; 
    [name] =&gt; named entity recognition
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; Named entity recognition is a form of {{entity recognition}} where the entities being detected have a proper name such a person's name or a placename.  Whilst datatyopes such as numbers and date may use some ofrm of {{pattern matching}} named entity recognition usually requires looking up individual words or {{n-grams}} in large lexicons or personal data sources.
)
  --></li>
  <li ><span class="name"><a href="https://alandix.com/glossary/aibook/named graphs">named graphs</a></span><!--  Array
(
    [slug] =&gt; named graphs
    [parent] =&gt; 
    [name] =&gt; named graphs
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; 
)
  --></li>
  <li ><span class="name"><a href="https://alandix.com/glossary/aibook/namespaces">namespaces</a></span><!--  Array
(
    [slug] =&gt; namespaces
    [parent] =&gt; 
    [name] =&gt; namespaces
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; 
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/aibook/national insurance number">National Insurance number</a></span><!--  Array
(
    [slug] =&gt; national insurance number
    [parent] =&gt; 
    [name] =&gt; National Insurance number
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; The National Insurance number (or NI number) is the principal {{personal identifier}} used in the UK for government purposes.  There are separate identifiers for health and tax.
)
  --></li>
  <li ><span class="name"><a href="https://alandix.com/glossary/aibook/natural language database query">natural language database query</a></span><!--  Array
(
    [slug] =&gt; natural language database query
    [parent] =&gt; 
    [name] =&gt; natural language database query
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; 
)
  --></li>
  <li ><span class="name"><a href="https://alandix.com/glossary/aibook/natural language processing">natural language processing</a></span>- has subentries<ul class="subentries"><li ><span class="name">database access</span> - See <span ><a href="https://alandix.com/glossary/aibook/natural language database query">natural language database query</a></span><!--  Array
(
    [slug] =&gt; natural language processing!database access
    [parent] =&gt; natural language processing
    [name] =&gt; database access
    [canonical] =&gt; 
    [see] =&gt; natural language database query
    [description] =&gt; 
)
  --></li>
<li ><span class="name">database queries</span> - See <span ><a href="https://alandix.com/glossary/aibook/natural language database query">natural language database query</a></span><!--  Array
(
    [slug] =&gt; natural language processing!database queries
    [parent] =&gt; natural language processing
    [name] =&gt; database queries
    [canonical] =&gt; 
    [see] =&gt; natural language database query
    [description] =&gt; 
)
  --></li>
<li ><span class="name"><a href="https://alandix.com/glossary/aibook/natural language processing!information management">information management</a></span><!--  Array
(
    [slug] =&gt; natural language processing!information management
    [parent] =&gt; natural language processing
    [name] =&gt; information management
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; 
)
  --></li>
<li ><span class="name"><a href="https://alandix.com/glossary/aibook/natural language processing!interfaces">interfaces</a></span><!--  Array
(
    [slug] =&gt; natural language processing!interfaces
    [parent] =&gt; natural language processing
    [name] =&gt; interfaces
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; 
)
  --></li>
<li ><span class="name"><a href="https://alandix.com/glossary/aibook/natural language processing!tagger">tagger</a></span><!--  Array
(
    [slug] =&gt; natural language processing!tagger
    [parent] =&gt; natural language processing
    [name] =&gt; tagger
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; 
)
  --></li>
<li ><span class="name"><a href="https://alandix.com/glossary/aibook/natural language processing!tagset">tagset</a></span><!--  Array
(
    [slug] =&gt; natural language processing!tagset
    [parent] =&gt; natural language processing
    [name] =&gt; tagset
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; 
)
  --></li></ul><!--  Array
(
    [slug] =&gt; natural language processing
    [parent] =&gt; 
    [name] =&gt; natural language processing
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; 
    [subentries] =&gt; Array
        (
            [natural language processing!database access] =&gt; Array
                (
                    [slug] =&gt; natural language processing!database access
                    [parent] =&gt; natural language processing
                    [name] =&gt; database access
                    [canonical] =&gt; 
                    [see] =&gt; natural language database query
                    [description] =&gt; 
                )

            [natural language processing!database queries] =&gt; Array
                (
                    [slug] =&gt; natural language processing!database queries
                    [parent] =&gt; natural language processing
                    [name] =&gt; database queries
                    [canonical] =&gt; 
                    [see] =&gt; natural language database query
                    [description] =&gt; 
                )

            [natural language processing!information management] =&gt; Array
                (
                    [slug] =&gt; natural language processing!information management
                    [parent] =&gt; natural language processing
                    [name] =&gt; information management
                    [canonical] =&gt; 
                    [see] =&gt; 
                    [description] =&gt; 
                )

            [natural language processing!interfaces] =&gt; Array
                (
                    [slug] =&gt; natural language processing!interfaces
                    [parent] =&gt; natural language processing
                    [name] =&gt; interfaces
                    [canonical] =&gt; 
                    [see] =&gt; 
                    [description] =&gt; 
                )

            [natural language processing!tagger] =&gt; Array
                (
                    [slug] =&gt; natural language processing!tagger
                    [parent] =&gt; natural language processing
                    [name] =&gt; tagger
                    [canonical] =&gt; 
                    [see] =&gt; 
                    [description] =&gt; 
                )

            [natural language processing!tagset] =&gt; Array
                (
                    [slug] =&gt; natural language processing!tagset
                    [parent] =&gt; natural language processing
                    [name] =&gt; tagset
                    [canonical] =&gt; 
                    [see] =&gt; 
                    [description] =&gt; 
                )

        )

)
  --></li>
  <li ><span class="name"><a href="https://alandix.com/glossary/aibook/natural language understanding">natural language understanding</a></span><!--  Array
(
    [slug] =&gt; natural language understanding
    [parent] =&gt; 
    [name] =&gt; natural language understanding
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; 
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/aibook/natural logarithm">natural logarithm</a></span><!--  Array
(
    [slug] =&gt; natural logarithm
    [parent] =&gt; 
    [name] =&gt; natural logarithm
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; The natural logarithm is a {{logarithm}} using e (Euler's number ~ 2.72) as base, it is often denoted &quot;ln&quot;.  That is ln(x) is the natural logarithm of x, and x = e&lt;sup&gt;ln(x)&lt;/sup&gt;.  Other common {{logarithm bases}} are 2 and 10, which are easier to interpret.  However natural logairithms occur in many mathematical and physical processes.
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/aibook/natural selection">natural selection</a></span><!--  Array
(
    [slug] =&gt; natural selection
    [parent] =&gt; 
    [name] =&gt; natural selection
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; In biology {{natural selection}} refers to the way that indviduals who are more fit or able to cope in their envirnment are more likely to survive to breeding age and bring up offspring successfully ({{survival of the fitest}}).  This means that their {[genes}} propogate more to the next generation leading to a gradual shift of the population towards a more successful gene pool for the enviornment.  In the natural environment this process can happen relatvely quickly, espeically in fast breeding organisms such as insects or disease pathogens.  It is one of the processes behind {{Darwinian evolution}}, although the latter also requires genetic {{mutation}} to allow the gene pool to gradually shift also.  These processes are emulated in artifical {{genetic algorithms}}.
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/aibook/negation as failure">negation as failure</a></span><!--  Array
(
    [slug] =&gt; negation as failure
    [parent] =&gt; 
    [name] =&gt; negation as failure
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; Some implemetations of logic, notably {{Prolog}} apply negation as failure, whuch means that if a solution is not found (failure), the thing you are trying to solve is deemed as not true (negation).  This effectvely adopts the {{closed world assumption}} - if raw facts are not explciitly stated they are false, or in other words all facts are known in the {{knowledge base}}.
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/aibook/nettalk">NETtalk </a></span><!--  Array
(
    [slug] =&gt; nettalk
    [parent] =&gt; 
    [name] =&gt; NETtalk 
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; NETtalk was an early text to speech application built using a {{multi-layer perceptron}} built in the mid 1980s..  In some way NETtalk's training is very smilar to more recent systems, using data form surrounding letters to predict the ecorrect phoneme to pronounce.  However, in comparison the the many millions of neurons in current networks, NETtalk had a mere  203 input units, 80 hidden units and 26 output units. Yet despite this produced intelligable output and learning behavour rather like a child's.   Certainly a lesson for those today trying to reduce the {{carbon footprint}} of AI and reduce the {{digital divide}}.
)
  --></li>
  <li ><span class="name"><a href="https://alandix.com/glossary/aibook/network representation">network representation</a></span><!--  Array
(
    [slug] =&gt; network representation
    [parent] =&gt; 
    [name] =&gt; network representation
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; 
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/aibook/neural correlates of consciousness">neural correlates of consciousness</a></span><!--  Array
(
    [slug] =&gt; neural correlates of consciousness
    [parent] =&gt; 
    [name] =&gt; neural correlates of consciousness
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; Neural correlates of consciousness are processes and behaviours in the brain that occur at the same time as {{subjective}} {{conscious}} experience, for example, {{global synchronisation of oscillations}}  where distant parts of the brain fire in the same rhythm.  As the maxim goies &quot;{{correlation does not imply causation}}&quot;, so one cannot say that the experince consciousness .is because of the particular neural phenomenon, but it does help to sfta nd inspire more theoretical, philosophical and practical approaches to {{consciousness}}.
)
  --></li>
  <li ><span class="name"><a href="https://alandix.com/glossary/aibook/neural network">neural network</a></span><!--  Array
(
    [slug] =&gt; neural network
    [parent] =&gt; 
    [name] =&gt; neural network
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; 
)
  --></li>
  <li ><span class="name"><a href="https://alandix.com/glossary/aibook/neuron">neuron</a></span><!--  Array
(
    [slug] =&gt; neuron
    [parent] =&gt; 
    [name] =&gt; neuron
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; 
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/aibook/neurones">neurones</a></span><!--  Array
(
    [slug] =&gt; neurones
    [parent] =&gt; 
    [name] =&gt; neurones
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; Neurone is an alternative spelling of {{neuron}}.  Sometimes neurone, with the 'e' may be used to emphasise when the subjcet is bilogical rather than artificoal neurons.
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/aibook/neurosymbolc ai">neurosymbolc AI</a></span><!--  Array
(
    [slug] =&gt; neurosymbolc ai
    [parent] =&gt; 
    [name] =&gt; neurosymbolc AI
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; Neurosymbolc AI refers to techniques that combine {{sub-symbolic}} {{neural networks}} with {{symbolic}} techniques.  These may be fairly distinct modules in a {{hybrid archtecture}}, but also can be integrated at.a more fundamental way such as using neural techniques to reason over {{semantic networks}} or the way high-level {{counterfactual reasonng}} is used to guide low-level stimulus-response learning in the {{cognitive model of regret}}.
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/aibook/newwave interface">newwave interface</a></span><!--  Array
(
    [slug] =&gt; newwave interface
    [parent] =&gt; 
    [name] =&gt; newwave interface
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; NewWave was an alternative user interface for early versions of Windows on Hewlett-Packard hardware.  It included an {{intelligent agent}} portrayed as icon looking like a secret agent.
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/aibook/newell, alan">Newell, Alan</a></span><!--  Array
(
    [slug] =&gt; newell, alan
    [parent] =&gt; 
    [name] =&gt; Newell, Alan
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; Alan Newell was one of the early pioneers of AI amonst other things proposing (with {{Herbert Simon}}) the {{physical symbol system hypothesis}} and developing (with {{John Laird}} and {{Paul Rosenbloom}}) the {{SOAR}} cognitive architecture.
)
  --></li>
  <li ><span class="name">NLP</span> - See <span ><a href="https://alandix.com/glossary/aibook/natural language processing">natural language processing</a></span><!--  Array
(
    [slug] =&gt; nlp
    [parent] =&gt; 
    [name] =&gt; NLP
    [canonical] =&gt; 
    [see] =&gt; natural language processing
    [description] =&gt; 
)
  --></li>
  <li ><span class="name"><a href="https://alandix.com/glossary/aibook/non-linear transformations">non-linear transformations</a></span><!--  Array
(
    [slug] =&gt; non-linear transformations
    [parent] =&gt; 
    [name] =&gt; non-linear transformations
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; 
)
  --></li>
  <li ><span class="name">non-locality</span> - See <span class="has_content"><a href="https://alandix.com/glossary/aibook/locality">locality</a></span><!--  Array
(
    [slug] =&gt; non-locality
    [parent] =&gt; 
    [name] =&gt; non-locality
    [canonical] =&gt; 
    [see] =&gt; locality
    [description] =&gt; 
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/aibook/non-monotonic reasoning">non-monotonic reasoning</a></span><!--  Array
(
    [slug] =&gt; non-monotonic reasoning
    [parent] =&gt; 
    [name] =&gt; non-monotonic reasoning
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; In non-monotonic reasoning new knowledge can change whether somethibg is infered to be true.  Non-montinic reasing often uses {{default rules}}.  For example, if Hoppy is a dog, and we know that dogs (usually) have four legs, then we might conclude that the statement &quot;Hoppy has four legs&quot; is true.  However, if we later learn that Hoppy had had a car accident as a puppy and had a leg amputated, we would then need to change our reasoning and instead conclude that &quot;Hoppy has four legs&quot; is false. 
)
  --></li>
  <li ><span class="name"><a href="https://alandix.com/glossary/aibook/non-overlapping windows">non-overlapping windows</a></span><!--  Array
(
    [slug] =&gt; non-overlapping windows
    [parent] =&gt; 
    [name] =&gt; non-overlapping windows
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; 
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/aibook/non-terminal symbols">non-terminal symbols</a></span><!--  Array
(
    [slug] =&gt; non-terminal symbols
    [parent] =&gt; 
    [name] =&gt; non-terminal symbols
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; In a {{grammar}}, a non-terminal symbol is one that may be expanded into further symbols, for example a noun-phrase may expand to a determiner and a noun.  In terms of a {{parse tree}}, non-terminal symbols usually label nodes wirthin the body of the tree, whereas {{terminal symbols}} label the leaves.
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/aibook/non-zero-sum game">non-zero-sum game</a></span><!--  Array
(
    [slug] =&gt; non-zero-sum game
    [parent] =&gt; 
    [name] =&gt; non-zero-sum game
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; This is a game where one player's gain does not inevitably lead to the other player's loss.  This can be fully cooperative, for example, when several people need to solve parts of a larger puzzle, or more complicated, such as in the {{prisoners dilemma}}.  This is the opposite of a {{zero-sum game}}.
)
  --></li>
  <li ><span class="name"><a href="https://alandix.com/glossary/aibook/nonlinear planning">nonlinear planning</a></span><!--  Array
(
    [slug] =&gt; nonlinear planning
    [parent] =&gt; 
    [name] =&gt; nonlinear planning
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; 
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/aibook/normal distribution">Normal distribution</a></span><!--  Array
(
    [slug] =&gt; normal distribution
    [parent] =&gt; 
    [name] =&gt; Normal distribution
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; The Normal or Gaussian distribution is the bell-shaped {{probability distribution}} that is often found in natural phenomena such as weights and heights.  Some forms of {{statistical anlaysis}} may assume that data is Normally distributed.  Also some algorithms may use the Normal distribution as part of theor processing, for example {{Gaussian filters}} for {{smoothing}} an image.  Normal distributions are very common because they arise when large numbers of nearly {{independent}} small items are added tgeher (e.g. the effect of large numbers of genes and environmental effects).  Even though they are common, it is important to remember that there are also many kinds of {{non-Normal distrubution}} for example {{power-law distributions}} that occur in much web and social network data.
)
  --></li>
  <li ><span class="name">normalised information gain</span> - See <span class="has_content"><a href="https://alandix.com/glossary/aibook/information gain ratio">information gain ratio</a></span><!--  Array
(
    [slug] =&gt; normalised information gain
    [parent] =&gt; 
    [name] =&gt; normalised information gain
    [canonical] =&gt; 
    [see] =&gt; information gain ratio
    [description] =&gt; 
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/aibook/normalised mutual information">normalised mutual information</a></span><!--  Array
(
    [slug] =&gt; normalised mutual information
    [parent] =&gt; 
    [name] =&gt; normalised mutual information
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; Normalised mutual information is an {{information measure}} used to assess the quality of {{clustering}} and {{classification algorithms}} ({{concept learning}}) in general.  If you split classified data into several groups, then the {{entropy}} of the result will be higher, even if the division is entirely {{random}} and unconnected to the target classification, simply becasue the reulting groups and hence probabilities are smaller.          It is obtained by dividing the raw {{information gain}} (gain(C,G)) by the sum of the entropy of the classification and decision grouping.  That is &lt;br&gt; &amp;nbsp; Normalised information gain &amp;nbsp; =  &amp;nbsp; gain(C,G)  /  ( H(C) + H(G) ) &lt;br&gt; where H(G) is the entropy of the division into the groups. &lt;br&gt;&lt;br&gt;. This is similar to but different from {{normalised information gain}}, as the latter just divides by H(G).  In practice this means normalised mutual information it is weighted slightly less against larger numbers of groups compared with  {{normalised information gain}}.
)
  --></li>
  <li ><span class="name"><a href="https://alandix.com/glossary/aibook/normalization">normalization</a></span><!--  Array
(
    [slug] =&gt; normalization
    [parent] =&gt; 
    [name] =&gt; normalization
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; 
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/aibook/noughts and crosses">noughts and crosses</a></span><!--  Array
(
    [slug] =&gt; noughts and crosses
    [parent] =&gt; 
    [name] =&gt; noughts and crosses
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; The childs's game noughts and crosses (called tic-tac-toe in the USA) can be useful as. simple test case for computer {{game playing algorithms}}.  It is an example of a {{zero-sum game}}.
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/aibook/obfuscation">obfuscation</a></span><!--  Array
(
    [slug] =&gt; obfuscation
    [parent] =&gt; 
    [name] =&gt; obfuscation
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; Obfusication is when you try to conceal soemthing by burying it amongst other data or modifying it.  In particular {{privacy}} mechanisms may try to obfusicate perosnal details, for exmaple, slightly changing your birthday, or house number.
)
  --></li>
  <li ><span class="name"><a href="https://alandix.com/glossary/aibook/object identification">object identification</a></span>- has subentries<ul class="subentries"><li ><span class="name"><a href="https://alandix.com/glossary/aibook/object identification!bitmaps">bitmaps</a></span><!--  Array
(
    [slug] =&gt; object identification!bitmaps
    [parent] =&gt; object identification
    [name] =&gt; bitmaps
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; 
)
  --></li>
<li ><span class="name"><a href="https://alandix.com/glossary/aibook/object identification!outlines">outlines</a></span><!--  Array
(
    [slug] =&gt; object identification!outlines
    [parent] =&gt; object identification
    [name] =&gt; outlines
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; 
)
  --></li>
<li ><span class="name"><a href="https://alandix.com/glossary/aibook/object identification!paths">paths</a></span><!--  Array
(
    [slug] =&gt; object identification!paths
    [parent] =&gt; object identification
    [name] =&gt; paths
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; 
)
  --></li>
<li ><span class="name"><a href="https://alandix.com/glossary/aibook/object identification!summary statistics">summary statistics</a></span><!--  Array
(
    [slug] =&gt; object identification!summary statistics
    [parent] =&gt; object identification
    [name] =&gt; summary statistics
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; 
)
  --></li></ul><!--  Array
(
    [slug] =&gt; object identification
    [parent] =&gt; 
    [name] =&gt; object identification
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; 
    [subentries] =&gt; Array
        (
            [object identification!bitmaps] =&gt; Array
                (
                    [slug] =&gt; object identification!bitmaps
                    [parent] =&gt; object identification
                    [name] =&gt; bitmaps
                    [canonical] =&gt; 
                    [see] =&gt; 
                    [description] =&gt; 
                )

            [object identification!outlines] =&gt; Array
                (
                    [slug] =&gt; object identification!outlines
                    [parent] =&gt; object identification
                    [name] =&gt; outlines
                    [canonical] =&gt; 
                    [see] =&gt; 
                    [description] =&gt; 
                )

            [object identification!paths] =&gt; Array
                (
                    [slug] =&gt; object identification!paths
                    [parent] =&gt; object identification
                    [name] =&gt; paths
                    [canonical] =&gt; 
                    [see] =&gt; 
                    [description] =&gt; 
                )

            [object identification!summary statistics] =&gt; Array
                (
                    [slug] =&gt; object identification!summary statistics
                    [parent] =&gt; object identification
                    [name] =&gt; summary statistics
                    [canonical] =&gt; 
                    [see] =&gt; 
                    [description] =&gt; 
                )

        )

)
  --></li>
  <li ><span class="name"><a href="https://alandix.com/glossary/aibook/object recognition">object recognition</a></span><!--  Array
(
    [slug] =&gt; object recognition
    [parent] =&gt; 
    [name] =&gt; object recognition
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; 
)
  --></li>
  <li ><span class="name"><a href="https://alandix.com/glossary/aibook/object-oriented architecture">object-oriented architecture</a></span><!--  Array
(
    [slug] =&gt; object-oriented architecture
    [parent] =&gt; 
    [name] =&gt; object-oriented architecture
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; 
)
  --></li>
  <li ><span class="name"><a href="https://alandix.com/glossary/aibook/object-oriented programming">object-oriented programming</a></span><!--  Array
(
    [slug] =&gt; object-oriented programming
    [parent] =&gt; 
    [name] =&gt; object-oriented programming
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; 
)
  --></li>
  <li ><span class="name">objective function</span> - See <span class="has_content"><a href="https://alandix.com/glossary/aibook/fitness function">fitness function</a></span><!--  Array
(
    [slug] =&gt; objective function
    [parent] =&gt; 
    [name] =&gt; objective function
    [canonical] =&gt; 
    [see] =&gt; fitness function
    [description] =&gt; 
)
  --></li>
  <li ><span class="name"><a href="https://alandix.com/glossary/aibook/obstacle avoidance">obstacle avoidance</a></span><!--  Array
(
    [slug] =&gt; obstacle avoidance
    [parent] =&gt; 
    [name] =&gt; obstacle avoidance
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; 
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/aibook/occam's razor">Occam's razor</a></span><!--  Array
(
    [slug] =&gt; occam's razor
    [parent] =&gt; 
    [name] =&gt; Occam's razor
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; Occam's razor (variously spelt Ockam, Ocham) is the principle that the simplest explanation is often the best or correct one.  It traces back to a 13th century monk William of Ockham, and is often stated as in Latin as &quot;pluralitas non est ponenda sine necessitate&quot;, which trabslates to &ldquo;plurality (or entities) should not be posited without necessity.&rdquo;  In AI this principle is often used to improve the {{generalisation}} after {{machine learning}} , for example {{pruning decision trees}} or the {{pinch points}} in {{deep neural networks}}.  It is closely related to issues of {{overfitting}} as models with too many free parameters often pick up irrelevant details of particular training examples.
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/aibook/ocr">OCR (optical character recognition)</a></span><!--  Array
(
    [slug] =&gt; ocr
    [parent] =&gt; 
    [name] =&gt; OCR (optical character recognition)
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; OCR (optical character recognition) is usually the first computer process to understand written or printed text.  Traditional methods involve matching against known patterns for letters, or first lookoing for features such as loops and lines. Neural networks are now used expecially for hand-written characters where ther is more variation.  A problem for both is matching chracters at different orientations, so some systems try to work out the general direction of text first and rotate the image, some rotate individual characters to match templates (called {{normalization}}).  Some neural networks are oreintation-indpeendent, by having layers which effectively perform this normalization.   Usuually the result of OCR is uncertain, both because of inherent ambiguities (is it zero or capital O, a small or large letter O, a 6 or a 9 upside down, VV or W?) and because the text may be poorly written or handwriting unusual.  This is resolved at higher levels of processing, for example when characters are grouped into word units they can be checked against a dictionary, allowing {{disambiguation}} of the lower level.  Typically some of this higher level processing is carried but within OCR resulting in plain text, but in some applications more raw confidence measuers are useful to allow yet higher-level dismbiguation, or highlighting of regions for hand checking.  
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/aibook/octree">octree</a></span><!--  Array
(
    [slug] =&gt; octree
    [parent] =&gt; 
    [name] =&gt; octree
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; Octrees are the 3D version of {{quadtrees}}.  The octree is a tree-based data representation to store and index location information about a 3D image or area.  The root of the tree has eight branches corresponding to the eight {{octants}} of the image/area.  For example, if the area is a unit square, then the first branch is for the area with coprdinates (0&ndash;0.5,0&ndash;0.5,0&ndash;0.5), and the next one (0&ndash;0.5,0&ndash;0.5,0.5&ndash;1).  If there are no data points in any of these octants, then they are marked as empty.  Otherwise the points are allocated to the relevant octant and then the same process is used to build an octree of those.
)
  --></li>
  <li ><span class="name"><a href="https://alandix.com/glossary/aibook/oncue">oncue</a></span><!--  Array
(
    [slug] =&gt; oncue
    [parent] =&gt; 
    [name] =&gt; oncue
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; 
)
  --></li>
  <li ><span class="name"><a href="https://alandix.com/glossary/aibook/ontology">ontology</a></span><!--  Array
(
    [slug] =&gt; ontology
    [parent] =&gt; 
    [name] =&gt; ontology
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; 
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/aibook/openai">OpenAI</a></span><!--  Array
(
    [slug] =&gt; openai
    [parent] =&gt; 
    [name] =&gt; OpenAI
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; Open AI is the company behind the {{large language models}} {{GPT-3}} and {{GPT-4}} that power the {{chatbit}} {{chatGPT}}.  
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/aibook/open ai">open AI</a></span><!--  Array
(
    [slug] =&gt; open ai
    [parent] =&gt; 
    [name] =&gt; open AI
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; The term 'open AI' refers to AI technology and software that is open in the sense of {{open source}} or {{public domain}}.  This includes a variety of types of licence, and may refer to everything about a systme or jsut some aspects.  For example, one might make. a {{generative model}} available in the public domain, but not the training set that gave rise to it, nor the algorithms used to create it.  Not to be confused with the company {{OpenAI}}.
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/aibook/open list">open list</a></span><!--  Array
(
    [slug] =&gt; open list
    [parent] =&gt; 
    [name] =&gt; open list
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; In {{search algorithms}} the open list is tothe (usually ordered) list of nodes that still need to be explored.  The basic {{breadth-first}} and {{depth-first}} search simply take nodes in turn from the front of the open list to explore next, they differ only in whether the children of nodes that are explored are added to the front or end of the open list.  More sophisticated algorithms (such as {{A*}}) may use more complex rules to decide which item to take next from the open list, often using {{heuristics} to guide the choice.  {{Graph search}} algorithms often also require a {{closed list}} recording the nodes that have already been considered.
)
  --></li>
  <li ><span class="name"><a href="https://alandix.com/glossary/aibook/open world assumption">open world assumption</a></span><!--  Array
(
    [slug] =&gt; open world assumption
    [parent] =&gt; 
    [name] =&gt; open world assumption
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; 
)
  --></li>
  <li ><span class="name"><a href="https://alandix.com/glossary/aibook/open-loop control">open-loop control</a></span><!--  Array
(
    [slug] =&gt; open-loop control
    [parent] =&gt; 
    [name] =&gt; open-loop control
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; 
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/aibook/ops5">OPS5</a></span><!--  Array
(
    [slug] =&gt; ops5
    [parent] =&gt; 
    [name] =&gt; OPS5
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; OPS5 is an early {{production system}} language that was used in the {[expert system}} {{XCON}}.
)
  --></li>
  <li ><span class="name">optical character recognition</span> - See <span class="has_content"><a href="https://alandix.com/glossary/aibook/ocr">OCR (optical character recognition)</a></span><!--  Array
(
    [slug] =&gt; optical character recognition
    [parent] =&gt; 
    [name] =&gt; optical character recognition
    [canonical] =&gt; 
    [see] =&gt; ocr
    [description] =&gt; 
)
  --></li>
  <li ><span class="name"><a href="https://alandix.com/glossary/aibook/optical flow">optical flow</a></span><!--  Array
(
    [slug] =&gt; optical flow
    [parent] =&gt; 
    [name] =&gt; optical flow
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; 
)
  --></li>
  <li ><span class="name"><a href="https://alandix.com/glossary/aibook/optimal">optimal</a></span><!--  Array
(
    [slug] =&gt; optimal
    [parent] =&gt; 
    [name] =&gt; optimal
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; 
)
  --></li>
  <li ><span class="name"><a href="https://alandix.com/glossary/aibook/optimality">optimality</a></span><!--  Array
(
    [slug] =&gt; optimality
    [parent] =&gt; 
    [name] =&gt; optimality
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; 
)
  --></li>
  <li ><span class="name"><a href="https://alandix.com/glossary/aibook/optimization">optimization</a></span><!--  Array
(
    [slug] =&gt; optimization
    [parent] =&gt; 
    [name] =&gt; optimization
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; 
)
  --></li>
  <li ><span class="name"><a href="https://alandix.com/glossary/aibook/optional values">optional values</a></span><!--  Array
(
    [slug] =&gt; optional values
    [parent] =&gt; 
    [name] =&gt; optional values
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; 
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/aibook/outliers">outliers</a></span><!--  Array
(
    [slug] =&gt; outliers
    [parent] =&gt; 
    [name] =&gt; outliers
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; Outliers are data items that are very unusual, for example having particularly large or small values of a numerical feature.. Outliers can be problematic, especially for statistical methods that use the {{arithmetic mean}}, {{variance}} or similar measures, all of which are quite sensitive to outliers.  This can also be a problem in pre-provessing for AI and ML algorithms if, for example, we use the mid-point of the full range of a feature as a way of reducing it a categorical large--small derived feature. One way to avoid these problems is to use metrics, such as the {{median}} or {{interquartile range}}, or to {{pre-whiten}} data such as scaling it to {{percentiles}}.  Another approach is to remove outliers form the data, for example, a common rule is to remove items where data values are more than twice the {{standard devoation}}.  It is important to have a fixed, pre-determined rule for this if possible to avoid {{cherry picking}} the removal of outliers that simpyl give results you don't like, and hence adding your own {{bias}} to the data.
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/aibook/overfitting">overfitting</a></span><!--  Array
(
    [slug] =&gt; overfitting
    [parent] =&gt; 
    [name] =&gt; overfitting
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; Overfitting is when a {{machine learning}} or a {{statistical model}} starts to model very particular features of the {{training set}} rather than the more general features of the underlying populaton or phenomena the training set is intended to be representative of.  In the extreme the learning process precsiely mateches the training exmaples, but nothing else.  This can be intentional, as in the case of {{memoisation}}, but usually an accdient due to {{fitting}} too many {{model parameters}}.  In general there should be far fewer parameters (e.g. weights in a {{neural network}}) than features in the trainng data set.  For exmaple, if you have a thousand training items with ten features per item, then you should have far less than ten thousand (1000 items x 10 features per item) parameters.  By these metrics, some {{large language models}} appear to have too many parameters, and indeed do sometimes show features of overfitting such as directly reproducing passages of training text.  However, there are arguments that this is not so critical as early layers of {[deep neural networks}} are producing large amounts of {{non-linear diversity}}, which is then effectvely internally selected and pruned by later layers, especially at {{pinch points}}.
)
  --></li>
  <li ><span class="name"><a href="https://alandix.com/glossary/aibook/overlearning">overlearning</a></span><!--  Array
(
    [slug] =&gt; overlearning
    [parent] =&gt; 
    [name] =&gt; overlearning
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; 
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/aibook/ovid">Ovid</a></span><!--  Array
(
    [slug] =&gt; ovid
    [parent] =&gt; 
    [name] =&gt; Ovid
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; Ovid was a Roman poet, who, amongst other works, wrote about {{Pygmalion}} a sculptor who fell in love with a statue he had carved. The state was brought to life by Aphrodite at his request. The story is often used as a parallel for makers of {{artificial intelligence}} and {{robots}}, especially those espousing {{strong AI}}.
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/aibook/owl">OWL</a></span><!--  Array
(
    [slug] =&gt; owl
    [parent] =&gt; 
    [name] =&gt; OWL
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; OWL (Web Ontology Language) is a family of notations for expressing properties of an {{RDF}} {{ontology}}, which vary in expressive power.  The simplest, {{OWLlite}}, can express whether instances of a class should have particlar properties (say that animals have a number of legs) or that a class is a sub-class of another (e.g. Dalmation is a kind of dog).  More complex variants allow oen to declare, for example, that a class can only have a fixed set of instances, or to use levels of meta-modelling where classes can be treated as instances.  In general, simpler variants are easier to deal with and check computationally, but are less expressive.
)
  --></li>
  <li ><span class="name"><a href="https://alandix.com/glossary/aibook/pagerank">PageRank</a></span><!--  Array
(
    [slug] =&gt; pagerank
    [parent] =&gt; 
    [name] =&gt; PageRank
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; 
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/aibook/parallax">parallax</a></span><!--  Array
(
    [slug] =&gt; parallax
    [parent] =&gt; 
    [name] =&gt; parallax
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; Parallax is the way that the relative position of objects can appear different from different points of observation.  In particular for machine (and human) vision a more distant object may appear to the left of a closer object when viewed from your left eye and to the right when seen from your right eye, this allowing {{stereo vision}}.  It also is the reason that distant objects seem to follow you as you move, a principle used in some forms of {{active vision}} systems.
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/aibook/parallel processing">parallel processing</a></span><!--  Array
(
    [slug] =&gt; parallel processing
    [parent] =&gt; 
    [name] =&gt; parallel processing
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; Parallel processing is when several computers or computing elemnts within a single computer can run computation at the same time (in parallel).  It may occur at a fairly large scale, such as tow separate computers or two cores in a single CPU.  It may also occur at a fine scale, for example the way a {{GPU}} does the same operation to many streams of data similtaneously.   Parallel processing is often divided into two main types {{SIMD}} (single instruction, multiple data) when the same things happens to lots of data in parallel, as in the case of the {{GPU}}; and {{MIMD}} (multiple instruction, multiple data) when different code may operate at teh same time, as in the case of CPU cores.
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/aibook/pareto frontier">pareto frontier</a></span><!--  Array
(
    [slug] =&gt; pareto frontier
    [parent] =&gt; 
    [name] =&gt; pareto frontier
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; The Pareto frontier is the set of all Pareto-optimal solutions.  For example, in the figure below, S1, S4, S5 and S6 are on the Pareto frontier.  The Pareto-frontier may be useful in {{visualising}} the trade-offs in a {{multi-objective optimisation}}
)
  --></li>
  <li ><span class="name"><a href="https://alandix.com/glossary/aibook/pareto, vilfredo">Pareto, Vilfredo</a></span><!--  Array
(
    [slug] =&gt; pareto, vilfredo
    [parent] =&gt; 
    [name] =&gt; Pareto, Vilfredo
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; 
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/aibook/pareto-optimal">Pareto-optimal</a></span><!--  Array
(
    [slug] =&gt; pareto-optimal
    [parent] =&gt; 
    [name] =&gt; Pareto-optimal
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; In {{multi-objective optimisation}}, a value is said to be Pareto-optimal is there is no alternative value that is better for all criteria.  Typically there may be values where one is better on one criterion and the other on another.  For example, when considering ways to read, we may think paper books are best for ease of thumbing through pages rapidly, but that ebooks are best when searching.  There is no technology that is best on all criteria, but both paper books and ebooks are pareto-optimal, as neither is better than the other on all citeria.
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/aibook/parse tree">parse tree</a></span><!--  Array
(
    [slug] =&gt; parse tree
    [parent] =&gt; 
    [name] =&gt; parse tree
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; A parse tree is often the output of a {[parser}}.  It typically follows the {{grammar}} of the underlying natural language or formal notation, with higher level nodes tagged with {{non-terminal symbols}} of the {{grammar}} and leaves tagged with {{terminal symbols}} and connected to actual words in the text.  In a formal gramer, such as for programming language a mathematical formulae, the parse tree may be  executes or evaluated dircetly by an interpreter or compiled into lower-level code or later execution.
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/aibook/parser">parser</a></span><!--  Array
(
    [slug] =&gt; parser
    [parent] =&gt; 
    [name] =&gt; parser
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; The term 'parser' can be used in several, related, senses.  In {{natural language processing}} a parser is the code  that execute during {{syntactic analysis}} to match the {{gramamr}} of the language.  The term is also used in {{pattern matching}} systems, for example, in {{entity recognition}} used in {{data preparation}} or {{data detectors}} to match particular kinds of data such as serial numbers, addresses or formulae.  It can also be used in compilers and interpreters of programming languages when matching code to the formal grammar of the language.
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/aibook/part-of-speech tagger">part-of-speech tagger</a></span><!--  Array
(
    [slug] =&gt; part-of-speech tagger
    [parent] =&gt; 
    [name] =&gt; part-of-speech tagger
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; A part-of-speach tagger takes text and annotates it with tags representing parts of speach such as 'noun phrase' or 'adjective'.  It is often used as an early stage of {{natural language pricessing}} before {{syntactic analysis}} and further stages.
)
  --></li>
  <li ><span class="name"><a href="https://alandix.com/glossary/aibook/partial state">partial state</a></span><!--  Array
(
    [slug] =&gt; partial state
    [parent] =&gt; 
    [name] =&gt; partial state
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; 
)
  --></li>
  <li ><span class="name"><a href="https://alandix.com/glossary/aibook/pattern-based recognition">pattern-based recognition</a></span><!--  Array
(
    [slug] =&gt; pattern-based recognition
    [parent] =&gt; 
    [name] =&gt; pattern-based recognition
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; 
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/aibook/pen-based systems">pen-based systems</a></span><!--  Array
(
    [slug] =&gt; pen-based systems
    [parent] =&gt; 
    [name] =&gt; pen-based systems
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; Pen-based systems refers to devices that use a pen or stylus for input.  They usually require some form of {{handwriting recognition}} for text input and sometimes {{gesture recognition}} for actions such as deleting text.
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/aibook/penrose, roger">Penrose, Roger</a></span><!--  Array
(
    [slug] =&gt; penrose, roger
    [parent] =&gt; 
    [name] =&gt; Penrose, Roger
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; Roger Penrose is a physicist famous for his Nobel Prize winning work on black holes and also author on the mechanisms and philosophy of {{consciosness}}.
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/aibook/penumbra">penumbra</a></span><!--  Array
(
    [slug] =&gt; penumbra
    [parent] =&gt; 
    [name] =&gt; penumbra
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; The penumbra is the area under part shadow, typically around the edge of the full shadow.  In sunlight these are points on the ground where part of the sun is visble.  Metaphorically, we can think of data items close to  {{decison boundary}} as being in the penumbra, as small changes to a few features might change the classification.
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/aibook/percentiles">percentiles</a></span><!--  Array
(
    [slug] =&gt; percentiles
    [parent] =&gt; 
    [name] =&gt; percentiles
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; Percentiles are the pnts in a set of numerical values corresponding to how far (as a percentage) they sit through the ordered lost.  For example, the 10 percentile is the point such that 10% of the data values are below and 90% above.  There are special cases such as the {{median}} which is at the half way point, or 50 percentile, and the {{lower quartile}} and {{upper quartile}} that lie at the 25% and 75% poinrs respectively.  It can be useful to transform data into percentiles during {{data preparation}} as it effectively turns the data into a {{uniform distribution}} and reduces the undue influence of {{outliers}}.  It can also be useful for {{long-tail distributions}} as it reduces the impact of the small number of exterme tail values.
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/aibook/perception">perception</a></span><!--  Array
(
    [slug] =&gt; perception
    [parent] =&gt; 
    [name] =&gt; perception
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; Perception refers to the process and experience by which the raw {{sensation}} of our senses (such as sight and sound) becomon meaningful.  AI tries to emulate these proceses in {{computer vision}} and {{natural languag eprocessing}} as low-level processing is sued to drive more {{semantic}} understanding.  While it is easiest to thnk of a bottom up process from raw {{sensation}} to perception, in fact in human cognituon there are top-down processes at work, we see what we expewct to see, which can aid disambiguation, but also can have problems if high-level expectations become locked in.  Computer perception also is most successful when there is a mix of bottom-up and top-down processes.
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/aibook/perceptron">perceptron</a></span><!--  Array
(
    [slug] =&gt; perceptron
    [parent] =&gt; 
    [name] =&gt; perceptron
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; The perceptron is the earliest kind of {{artificial neuron}}.  It has a number of inputs and weights for each.  In operation the iputs are multipled by the weights and summed to gib and input {{caruvation}}.  This is then compared with a {{threshold function}}, and if the input activation exceeds the threshold, the perceptron fires. A simple percepttron arranged ina. single layer acnnot solve complex problems such as the {{XOR problem}}, but multi-layer perceptrons are hard to train.  For many years this limited practicla developemnt of {{arificial neural networks}} until the adoption of {{non-linear threshold functons}} and the development {{backpropogation}}.
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/aibook/pereira, fernando">Pereira, Fernando</a></span><!--  Array
(
    [slug] =&gt; pereira, fernando
    [parent] =&gt; 
    [name] =&gt; Pereira, Fernando
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; Fernando Pereira is an early AI researcher, who amongst other things, was one of the creators of \term{definite clause grammar}, and is now VP and Engineering Fellow at Google.
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/aibook/personal information management">personal information management</a></span><!--  Array
(
    [slug] =&gt; personal information management
    [parent] =&gt; 
    [name] =&gt; personal information management
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; Personal information management (PIM) is the study of the way people gather, organise, retreive and reuse their own information.  It is usually in contrast to organisational data, such as large databases or customer relations managment (CRM) software, as PIM is concerned more with small data, often of an informal or individual nature.  However, PIM software may be used by people in organisations, for example, contact lists, as well as in domestic settings, for example budget planners.  PIM has often been the target application for work on {{intelligent agents}}.
)
  --></li>
  <li ><span class="name">personal knowledge management</span> - See <span class="has_content"><a href="https://alandix.com/glossary/aibook/personal information management">personal information management</a></span><!--  Array
(
    [slug] =&gt; personal knowledge management
    [parent] =&gt; 
    [name] =&gt; personal knowledge management
    [canonical] =&gt; 
    [see] =&gt; personal information management
    [description] =&gt; 
)
  --></li>
  <li ><span class="name">perturbation</span> - See <span class="has_content"><a href="https://alandix.com/glossary/aibook/perturbation techniques">perturbation techniques</a></span><!--  Array
(
    [slug] =&gt; perturbation
    [parent] =&gt; 
    [name] =&gt; perturbation
    [canonical] =&gt; 
    [see] =&gt; perturbation techniques
    [description] =&gt; 
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/aibook/perturbation techniques">perturbation techniques</a></span><!--  Array
(
    [slug] =&gt; perturbation techniques
    [parent] =&gt; 
    [name] =&gt; perturbation techniques
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; Perturbation techniques make small changes to inputs to a model or algorithm and the. track how they change the output.  They are {{black-box methods}} in that they do not need to know anything about the internal workings of the model or algorithm.  Perturbatuon techniques can be used to measure the {[sensitivity}} of a model to changes and hence the {{uncertainty}} of the model.  They can also be really useful for {{explainable AI}}, particularly when applied to provide a ({local explanation}} for a specific decision.
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/aibook/phase">phase</a></span><!--  Array
(
    [slug] =&gt; phase
    [parent] =&gt; 
    [name] =&gt; phase
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; In waves and frequency-based analaysis, such as {{Fourier transforms}}, phase refers to the time offset of the wave as opposed to the {{wavelength}} (how often it movs form higb to low and back) and the {{amplitude}} how high or low it goes.
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/aibook/phenomenological">phenomenological</a></span><!--  Array
(
    [slug] =&gt; phenomenological
    [parent] =&gt; 
    [name] =&gt; phenomenological
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; Phenomenological approaches take into account {{subjective experience}} as well as external behaviour.  For example, the experience of seeing red, not merely the 700 nanometre wavelength that excites the red-sensing cones in the human eye.
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/aibook/phenotype">phenotype</a></span><!--  Array
(
    [slug] =&gt; phenotype
    [parent] =&gt; 
    [name] =&gt; phenotype
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; In biological {{genetics}} the pheonotype is the observable characteristics of the organism in contrast to the {{genotype}} which is the actual gene structure that encodes it.  In computational {{genetic algorithms}} there is a similar difference between the way in which the state is encoded, for example, the locations of appliances ina. kitchen, and the way that gives rise to, say, plumbing and wiring issues within the kitchen.  Often the mapping between {{genotype}} and pheonotype is effectively a one-way function, where it is hard to work out the geotype given the phenotype.  However, for a {{genetic algorithm}} to work it is essential that there is at least some pattern to the relationship otherwise it will be no better than a {{Monte Carlo algorithm}}.
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/aibook/phone slave">Phone Slave</a></span><!--  Array
(
    [slug] =&gt; phone slave
    [parent] =&gt; 
    [name] =&gt; Phone Slave
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; Phone Slave was an early envisionment that used state of the art {{speech processing}} in the 1980s to show how an telephone answering machine could be transformed.  It prefigured many of the issues of {{speech-based interfaces}} such as how to avoid accidentally invoking commands when talking to other people in the environment.
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/aibook/phonetic typewriter">phonetic typewriter</a></span><!--  Array
(
    [slug] =&gt; phonetic typewriter
    [parent] =&gt; 
    [name] =&gt; phonetic typewriter
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; The phonetic typewriter was an early example of the use of {{Kohonen nets}}.
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/aibook/physical symbol system hypothesis">physical symbol system hypothesis</a></span><!--  Array
(
    [slug] =&gt; physical symbol system hypothesis
    [parent] =&gt; 
    [name] =&gt; physical symbol system hypothesis
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; The physical symbol system hypothesis, forst stated by {{Alan Newell}} in 1976, states that &quot;a physical symbol system has the necessary and sufficient means for general intelligent
action&quot;.  A physical symbol system in this sense comprises a set of tokens or symbols with rules to manipulate them.  In computer terms we can think of digital representatiosm, but the term can also be applied to physical tokens, as envisaged in {{Searle's Chinese room}}.  It is related to notions of {{physicalism}} in the {{philosophy of intelligence}} and {{consciousness}}, but is sometimes given a tighter meaning assumes a higher-level symbol system as found in {{symbolic AI}} as opposed to {{subsymbolic}} {{neural networks}}; however the altter can be argued to also be manioulating 'symbols' on a broad sense of the term, just vast numbers of them.
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/aibook/physicalism">physicalism</a></span><!--  Array
(
    [slug] =&gt; physicalism
    [parent] =&gt; 
    [name] =&gt; physicalism
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; Physicalism is the {{philosophical}} tenet that everything can be explained by physical processes.  This isthe domonant view in modern {{philosophy}} and {{physics}}.  This is closely associated with views that all intelligence and even consciousness can be explained by neurones in the brain acting as a form of {{information processing}} engine, and one of the arguments for {{strong AI}},   Physicalism is also losely related to the {{physical symbol system hypothesis}}, although the latter is often not taken to include {[sub-symbolic}} processing.    Within philosophy, physicalism and  {{materialism}} may be used interchangably, but the latter may sometimes be taken in a slightly more restricted sense of describing things interms of matter not including other aspects of the physical world such as energy fields.
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/aibook/piecewise linear">piecewise linear</a></span><!--  Array
(
    [slug] =&gt; piecewise linear
    [parent] =&gt; 
    [name] =&gt; piecewise linear
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; A piecewise linear fit or model is one that uses a collection of linear patches (in 2D straight lines) to approximate a curved line or (hyper)surface.
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/aibook/pinch-point layer">pinch-point layer</a></span><!--  Array
(
    [slug] =&gt; pinch-point layer
    [parent] =&gt; 
    [name] =&gt; pinch-point layer
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; Often a {{deep neural network}} includes one or more layers with far fewer nodes than others.  These pinch points often represent compact representations of {{features}}.  Thery may also be useful as part of techniques for {{explainable AI}}.
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/aibook/placing dominoes">placing dominoes</a></span><!--  Array
(
    [slug] =&gt; placing dominoes
    [parent] =&gt; 
    [name] =&gt; placing dominoes
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; Placing domonoes is a simple (and rahter boring) game played on a squared board with {{dominoes}} to illustrate {{game trees}}.
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/aibook/plateau">plateau</a></span><!--  Array
(
    [slug] =&gt; plateau
    [parent] =&gt; 
    [name] =&gt; plateau
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; A pleateau in a {{fitness landscape}} is a region where there is little or no change on the {{fitness function}}.  This can a problem for methdos that use any form of {{hill climbing}} as small changes make no difference and hence it is not clear what changes to meke in order to improve the fitness.
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/aibook/pogoing robots">Pogoing robots</a></span><!--  Array
(
    [slug] =&gt; pogoing robots
    [parent] =&gt; 
    [name] =&gt; Pogoing robots
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; Robots that hop on a single leg, like a child on a Pogo stick, are not very practically useful, but are a good test of control algorithms.  In particular, often {{fuzzy algorithms}} perform better than those trying work out the full dynamics.
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/aibook/pole balancing">pole balancing</a></span><!--  Array
(
    [slug] =&gt; pole balancing
    [parent] =&gt; 
    [name] =&gt; pole balancing
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; A simple control task is to move a trolly back and forth while balancing a pole on top.  This is not practically useful, but helps test different control strategies.  In particular, it has been used as a test case for {{reinforcement learning}}.
)
  --></li>
  <li ><span class="name"><a href="https://alandix.com/glossary/aibook/policy network">policy network</a></span><!--  Array
(
    [slug] =&gt; policy network
    [parent] =&gt; 
    [name] =&gt; policy network
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; 
)
  --></li>
  <li ><span class="name"><a href="https://alandix.com/glossary/aibook/poorly constrained">poorly constrained</a></span><!--  Array
(
    [slug] =&gt; poorly constrained
    [parent] =&gt; 
    [name] =&gt; poorly constrained
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; 
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/aibook/poplog">Poplog</a></span><!--  Array
(
    [slug] =&gt; poplog
    [parent] =&gt; 
    [name] =&gt; Poplog
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; Poplog is multi-paradigm language allowing both logic and impertative styles of programming in the same system.  It was developed in the early 1980s as an implementation of {{Prolog}} in POP-11, and is still under ongoing development.
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/aibook/position independent">position independent</a></span><!--  Array
(
    [slug] =&gt; position independent
    [parent] =&gt; 
    [name] =&gt; position independent
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; In {{computer vision}} and {{image processing}}, we say a method to recognise an object in an image is position independent if it works no matter where the target is within the image.  This may be achieved by training on lots of different offsets of the same images, by centering images on pertinent features, or by using models that are intrinsically position independent.  In additoon we often seek {{scale independence}} and {{rotational independence}} (it doesnt matter how big or what anfgle the object is). 
)
  --></li>
  <li ><span class="name"><a href="https://alandix.com/glossary/aibook/power law distribution">power law distribution</a></span><!--  Array
(
    [slug] =&gt; power law distribution
    [parent] =&gt; 
    [name] =&gt; power law distribution
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; 
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/aibook/pragmatic analysis">pragmatic analysis</a></span><!--  Array
(
    [slug] =&gt; pragmatic analysis
    [parent] =&gt; 
    [name] =&gt; pragmatic analysis
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; Pragmatic analysis is the final stage in {{natural language processing}} which tries to make sense of an utterance in {{context}}.  This may in part be used to disambiguate an utterance, for example, the word 'America' may mena the whole continant, North America as a whole or the USA depending on the speaker and context.   This can also be important for understanding what an utterance is &lt;em&gt;for&lt;/em&gt;.  For example, the statement &quot;it is cold&quot; can be interpretted as a simple fact about the temperature in the room, or as a {{speech act}} that means &quot;please shut the door&quot;.   Pragmatic analysis may follow {{semantic analysis}} in a {{natural language processing}} process, but also may be addeessed directly, for example by looking for particular patterns or terms in text.
)
  --></li>
  <li ><span class="name"><a href="https://alandix.com/glossary/aibook/pre-processing">pre-processing</a></span><!--  Array
(
    [slug] =&gt; pre-processing
    [parent] =&gt; 
    [name] =&gt; pre-processing
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; 
)
  --></li>
  <li ><span class="name"><a href="https://alandix.com/glossary/aibook/pre-processor">pre-processor</a></span><!--  Array
(
    [slug] =&gt; pre-processor
    [parent] =&gt; 
    [name] =&gt; pre-processor
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; 
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/aibook/pre-whitening">pre-whitening</a></span><!--  Array
(
    [slug] =&gt; pre-whitening
    [parent] =&gt; 
    [name] =&gt; pre-whitening
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; Often data values are transformed to make them more like a standard distribution. This process is called pre-whetning, especially for temporal data when the goal is to transform the data to be more like {{white noise}}.  Other examples are transforming data to make it more like a {{normal distribution}} of {{log transforms}} of timing data.
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/aibook/precision">precision</a></span><!--  Array
(
    [slug] =&gt; precision
    [parent] =&gt; 
    [name] =&gt; precision
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; When making a binary classification or decision, such as a disease diagnosis, precision is the {{probability}} that if we make a positive decision, it turns out out to be correct, for example, if we diagnose a person with the disease they really have it.  If a decision has a high precision, then the probability of a {{false positive}} is low.  Often increasing precision reduces {{recall}}; this is called the {{precision-recall trade-off}}.
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/aibook/precision-recall trade-off">precision&ndash;recall trade-off</a></span><!--  Array
(
    [slug] =&gt; precision-recall trade-off
    [parent] =&gt; 
    [name] =&gt; precision&amp;ndash;recall trade-off
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; Often algorithms give some sort of confidence measure for making a desicison, for example, a cancer diagnois.   We can then make the decision based on a threshold value, making a postive diagnosis if the confidence exceeds the threshold..  If the threshold of confidence is very high, we ensure high {{precision}}, that is if we make diagnosis, then it is almost certainyl right; but this then means that we are likley to miss some potential cases ({{false positives}}), that is low {{recall}}.  Alternateively a low threshold value will give high {[recall}}, we are unlikely to miss positive cases, but more likely to falsely diagnose someone ({{false negative}}), that is low {{precision}}.  In general, there is a trade-off between {{precision}} and {{recall}} and the correct balance is a {{cost--benefit}} trade-off depending on the relative impact of {{false positives}} and {{false negatives}}.
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/aibook/predicate logic">predicate logic</a></span><!--  Array
(
    [slug] =&gt; predicate logic
    [parent] =&gt; 
    [name] =&gt; predicate logic
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; Predicate logic (also known as first-order logic or precdiate caculus) allows logical formulae including variables (such as X, Y, Person, Parent)  and predicates (such as is_child_of(Person, Parent).  It does not allow higher-order formulae, that is predicte symbols cannot also be used as variables.
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/aibook/principal components analysis">principal components analysis</a></span><!--  Array
(
    [slug] =&gt; principal components analysis
    [parent] =&gt; 
    [name] =&gt; principal components analysis
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; Principal components analysis attempts to find a smaller dimensional subspace that captures the majority of the varation in a dataset.  It is a {{statistical technique}} that us used in various ways in AI.  Witin {{explainable AI}} it may be used to help interpret complex data, such as internal layers of a {[deep neural network}}.  It can also be used as a {{dimensional reduction}} technique during {{data preparation}}.  One way to calculate the principal components of a data set is first to calcualte the feature cross-correltion matrix, that is the {{matrix}} comprising of the {{correlations}} between different features.  The principal components are then the N {{eigenvectors}} with the largest {{eignevalues}}.   In. particular, the first principal component is the {{principal eigenvector}}. One can decide in advance how many principal components are required, or inspect them one by one choosing a point to stop where the eigenvalues drop below a chosen level. 
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/aibook/principal eigenvector">principal eigenvector</a></span><!--  Array
(
    [slug] =&gt; principal eigenvector
    [parent] =&gt; 
    [name] =&gt; principal eigenvector
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; The principal eigenvector is the {{eigenvector}} with the largest {[eigenvalue}}.  It is the direction that has the greatest variation within the datset.
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/aibook/prisoners' dilemma">prisoners' dilemma</a></span><!--  Array
(
    [slug] =&gt; prisoners' dilemma
    [parent] =&gt; 
    [name] =&gt; prisoners' dilemma
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; The prisoner's dilemma is a classic {{game theory}} problem.  Two prisoners are being separately interviewed by the police.  If one confeses and the other does not they one who confesses gets a reduced sentance of 1 year and the other gets a very long sentence of 10 years, if both confess they both get 5 years, but if neither confess they are freed.  The most risk-free choice for each prisoner is to confess (worst case 5 years), but the best thing is for neither to confess.  This is an example of a {{non-zero-sum game}} as the rewards/penalties are not a strict trade-off between the parties. It also shows the importance of communication.
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/aibook/privacy">privacy</a></span><!--  Array
(
    [slug] =&gt; privacy
    [parent] =&gt; 
    [name] =&gt; privacy
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; Privacy is about the ability or right of people to control what information or physical knowledge about themselves is available to others.  More broadly it is associated with {{personal integrity}} &ndash;&ndash; being in control of ones own body, mind and social relationships.  Many countries have {[data protection}} legislation which restricts those collecting or holding data.  Digital systems create threwats to privacy both through the accidental leakage of information or through deliberate or ill-considered sharing by others, such as personal photographs.  The harms can be financial (e.g. {{identity theft}}), social (e.g. embarassment), and psychological (e.g. feeling exposed).  Often provacy is seen purely as a form of {{secrecy}} of personal information, but the invasion of {{personal integrity}} is broader: for example {{deep fakes}} that create nude or other forms of false appearance or the use of aggregated data for purposes that the data subject would not have assented to.
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/aibook/privacy preserving algorithms">privacy preserving algorithms</a></span><!--  Array
(
    [slug] =&gt; privacy preserving algorithms
    [parent] =&gt; 
    [name] =&gt; privacy preserving algorithms
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; Privacy preserving algorithms attempt to perform machine learning on personal data without transfering the raw information to a central location, usually by processing locally on a user's own machine.  For example, deep neural networks make small differnce the the internal weights based on each training example.  Rather than send all of the training examples to a central location for processing, each person's local computer can work out the changes in the weights (known as a {{delta}}) and send these deltas where they can be combined to give overall learning.  It is hard to ensure total {{privacy}} as in some circumstances it has been shown that the original data can be reccovered from deltas.
)
  --></li>
  <li ><span class="name"><a href="https://alandix.com/glossary/aibook/probabilistic approaches">probabilistic approaches</a></span><!--  Array
(
    [slug] =&gt; probabilistic approaches
    [parent] =&gt; 
    [name] =&gt; probabilistic approaches
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; 
)
  --></li>
  <li ><span class="name"><a href="https://alandix.com/glossary/aibook/probabilistic process">probabilistic process</a></span><!--  Array
(
    [slug] =&gt; probabilistic process
    [parent] =&gt; 
    [name] =&gt; probabilistic process
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; 
)
  --></li>
  <li ><span class="name"><a href="https://alandix.com/glossary/aibook/probabilistic reasoning">probabilistic reasoning</a></span><!--  Array
(
    [slug] =&gt; probabilistic reasoning
    [parent] =&gt; 
    [name] =&gt; probabilistic reasoning
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; 
)
  --></li>
  <li ><span class="name"><a href="https://alandix.com/glossary/aibook/probability">probability</a></span><!--  Array
(
    [slug] =&gt; probability
    [parent] =&gt; 
    [name] =&gt; probability
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; 
)
  --></li>
  <li ><span class="name"><a href="https://alandix.com/glossary/aibook/probability density">probability density</a></span><!--  Array
(
    [slug] =&gt; probability density
    [parent] =&gt; 
    [name] =&gt; probability density
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; 
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/aibook/procedural knowledge">procedural knowledge</a></span><!--  Array
(
    [slug] =&gt; procedural knowledge
    [parent] =&gt; 
    [name] =&gt; procedural knowledge
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; Procedural knowledge encodes some sort of process or algorithm to create an answer in contrast to {{decarative knowledge}}.  This distinction applies to {{human reasing}}, for example you might know how to add up large numbers on an abacus (procedural knowledge) wheras another person might simply remember that 12+23 is 35 ({{decarative knowledge}}).  It also applies to {{automated reasoning}} , for example the use of {{grammars}} and {{parsing}} in {{natural language processing}} (procedural knowledge) vs. looking things up in large databases ({{decarative knowledge}}). 
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/aibook/proceduralisation">proceduralisation</a></span><!--  Array
(
    [slug] =&gt; proceduralisation
    [parent] =&gt; 
    [name] =&gt; proceduralisation
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; Proceduralisation is the process through which animals and also machine learning takes common sequences of actions, whether physical or mental, and combines them into a fixed unit, which is then one of the building blocks for future learning and action.  For example, a human guitar player may spend some time learning anew finger sequence, but after a while it no longer feels like separate movements, but is a single practiced motif.  Similarly, an automated proof algorithm might repeatedly apply the same sequence of algebraic manipulations during proofs, and after a while combine this into a new proof operator.   The term also has a specific meaning in the {{ACT-R}} cognitive architecture.
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/aibook/production rules">production rules</a></span><!--  Array
(
    [slug] =&gt; production rules
    [parent] =&gt; 
    [name] =&gt; production rules
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; A production rule is an IF &amp;lt;condition&amp;gt; THEN &amp;lt;action&amp;gt; rule as found in a {{production system}}.
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/aibook/production system">production system</a></span>- has subentries<ul class="subentries"><li ><span class="name">loans example</span> - See <span class="has_content"><a href="https://alandix.com/glossary/aibook/loans example">loans example</a></span><!--  Array
(
    [slug] =&gt; production system!loans example
    [parent] =&gt; production system
    [name] =&gt; loans example
    [canonical] =&gt; 
    [see] =&gt; loans example
    [description] =&gt; 
)
  --></li></ul><!--  Array
(
    [slug] =&gt; production system
    [parent] =&gt; 
    [name] =&gt; production system
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; A production system consists of three elements: (1). a database or working memory of facts, which can include temporary states such as &quot;Goal(open door)&quot;; (2) a set of ules of the form IF &amp;lt;condition&amp;gt; THEN &amp;lt;action&amp;gt;, where conditions are logical expressions about the working memory, and actions may include changes to the working memory such as &quot;ADD Goal(find key)&quot;; and (3) an interpretter, which matches the conditoons against the working memory and then executes the action for a matching rule.
    [subentries] =&gt; Array
        (
            [production system!loans example] =&gt; Array
                (
                    [slug] =&gt; production system!loans example
                    [parent] =&gt; production system
                    [name] =&gt; loans example
                    [canonical] =&gt; 
                    [see] =&gt; loans example
                    [description] =&gt; 
                )

        )

)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/aibook/programmable user models">programmable user models</a></span><!--  Array
(
    [slug] =&gt; programmable user models
    [parent] =&gt; 
    [name] =&gt; programmable user models
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; Programmable user models ({PUMS}} was a class of {{cognitive models}} based on the {{SOAR}} cognitive architecture.  The model was used to act as a surrogate user in testing of {{user interfaces}} or other user tasks such as arithmatic.  The purpose of PUMS was to emulate the user clsoely enough to be able to test deisgns of {{user interfaces}}, not to do the tasks to replace the user.
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/aibook/programming-by-demonstration">programming-by-demonstration</a></span><!--  Array
(
    [slug] =&gt; programming-by-demonstration
    [parent] =&gt; 
    [name] =&gt; programming-by-demonstration
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; While the term 'programming by demonstration' sounds as though it os about coding, in fac it was terned to desceribe systems that learn users tasks, notably the early {{intlligent agent}} {{EAGER}}.  A programming by demonstration system observes users actions and looks for patterns, repetations, etc. and then proposes ways to automate the task (the 'progamming' part of the name).
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/aibook/prolog">Prolog</a></span><!--  Array
(
    [slug] =&gt; prolog
    [parent] =&gt; 
    [name] =&gt; Prolog
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; Prolog is a prigramming language that is based on first order {{predicate logic}}.  It uses Horn clause logic (predicates can be defined in terms of other predicates, but not arbitrary equations) and {{unification}} of variables, with {{backtracking}} when it reaches an impasse or tries to seek additional solutions.  The outcome of a Prolog query is a binding of variables or 'No' meaning no (further) solution can be found.  Prolog allows negation and has a {{closed world}} model of knowledge -- if a fact is not found it is assumed to be false.
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/aibook/property inheritance">property inheritance</a></span><!--  Array
(
    [slug] =&gt; property inheritance
    [parent] =&gt; 
    [name] =&gt; property inheritance
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; Property inheritance is when a property of a class, such as &quot;dogs bark&quot; is assumed to be also true of sub-classes (such as Beagle) and instances (such as Snoopy).  This means that, in the absence of evidence to the contrary, we can infer that Beagles bark and that Snoopy barks.  
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/aibook/prospector">PROSPECTOR</a></span><!--  Array
(
    [slug] =&gt; prospector
    [parent] =&gt; 
    [name] =&gt; PROSPECTOR
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; PROSPECTOR was an expert system developed at Stanford in the
late 1970s  to evaluate geological sites for potential mineral deposits.  In 1984 it was
instrumental in discovering a molybdenum deposit worth 100 million dollars. &lt;br&gt;&lt;br&gt; PROSPECTOR  was based on a {{semantic network}} and used a combination of mostly {{forward chaining}} and some {{backward chaining}}.  It adopted a form of {{depth first search}} guided by the probabilities of different hypotheses and used {{Bayesian reasoning}}  to deal with uncertainty.  The {{dialogue component}} and {{explanation component}} used {{mixed control}}.
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/aibook/provenance">provenance</a></span><!--  Array
(
    [slug] =&gt; provenance
    [parent] =&gt; 
    [name] =&gt; provenance
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; Provenace is about the source of an object or information or the processes that created it.  For example, in the phsyical world, the provenance of an piece of antique furniture would include who designed and made it, who originally owned it, were it was bought and sold since.  Computationally, we may want to store this kind of information about a data item as part of its {{meta-information}}, and there are {{semantic web}} standards for storing provenance in {{RDF}}.  If the data item is refering to a phsyical thing (suc as antique furniture), then this may be exactly the kind of informationabove.  For other kinds of data it may include the source of the data, how it was collected or what datsource it came from; this can help assess the {{reliability}} of the data.  Outside the {{semantic web}}, during {{data preparation}} it is usually good to keep track of the processing steps that gave rise a to a particlar data file -- it is easy to forget!  Provevance can also be important in {{explainable AI}}, in order to increase the {{transparency}}  ofand hence {{trust}} in AI processes.
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/aibook/proxy indicator">proxy indicator</a></span><!--  Array
(
    [slug] =&gt; proxy indicator
    [parent] =&gt; 
    [name] =&gt; proxy indicator
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; When considering potemtial {{bias}}, a proxy indicator is some combination of features that reveal the value of a {{protected characteristic}} such as gender or ethnic background.  For example, Days taken as holiday from work might be indicative of a person's religion, or subjects tkane at school may be related to gender.  Often it is assumed that if {{protected characteristics}} are removed form a dataset before it is used for training, then the reuslying {{machine learning}} system will be free of {{bias}}.  However, proxy indicators mean that decisions are still effectievly made based on the {{protected characteristic}}.
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/aibook/pruning">pruning</a></span>- has subentries<ul class="subentries"><li ><span class="name">alpha&ndash;beta</span> - See <span class="has_content"><a href="https://alandix.com/glossary/aibook/alpha-beta pruning">alpha&ndash;beta pruning</a></span><!--  Array
(
    [slug] =&gt; pruning!alpha-beta
    [parent] =&gt; pruning
    [name] =&gt; alpha&amp;ndash;beta
    [canonical] =&gt; 
    [see] =&gt; alpha-beta pruning
    [description] =&gt; 
)
  --></li></ul><!--  Array
(
    [slug] =&gt; pruning
    [parent] =&gt; 
    [name] =&gt; pruning
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; Pruning is when one removes or ignores a part of a data structure. or process.  The term derives from horticultural pruning of dead or unwanted branches from a growing tree or bush.  Many of AI examples are about trees, on the computational sense, hence the use in AI.  One example of removal is to reduce the size of {{decison trees}} after and {{inductive learming}} algorithm such as {{ID3}}; pruning can remve branches that have low additional dscrininatiin and hence (typically) improve the {{generalisation}} of the tree.  An ignoring example is during {{A* search}} where the {{heuristic}} means one knows that a particular node cannot be on the best path, and hence can prune the node from the search of the graph.
    [subentries] =&gt; Array
        (
            [pruning!alpha-beta] =&gt; Array
                (
                    [slug] =&gt; pruning!alpha-beta
                    [parent] =&gt; pruning
                    [name] =&gt; alpha&amp;ndash;beta
                    [canonical] =&gt; 
                    [see] =&gt; alpha-beta pruning
                    [description] =&gt; 
                )

        )

)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/aibook/pseudonymisation">pseudonymisation</a></span><!--  Array
(
    [slug] =&gt; pseudonymisation
    [parent] =&gt; 
    [name] =&gt; pseudonymisation
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; When data is  {{pseudonymised}}, each person's name (or other personally identifiable labels) is replaced with a unique {{anonymous identifier}} .  This may be used when full {{anonymisation}} would ber too string.  For exampel, we may want to be able to reconnect processed data with an individual, perhaps to warn them of a disease.  It is also important when different datasets need to be linked together, we need to be able to tell which records refer to the same individual even if we do not know &lt;em&gt;which&lt;/em&gt; actual person is refered to.
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/aibook/pygmalion">Pygmalion</a></span><!--  Array
(
    [slug] =&gt; pygmalion
    [parent] =&gt; 
    [name] =&gt; Pygmalion
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; Pygmalion was a sculptor in the poem 'Metamorphoses' by the Roman poet {{Ovid}}. Pygmalion fell in love with a statue he had carved and asked Aphrodite to bring the statue to life.  However, once a flesh and blood women.  The story is often used as a parallel for makers of {{artificial intelligence}} and {{robots}}, esepcially those espousing {[strong AI}}.
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/aibook/python">Python</a></span><!--  Array
(
    [slug] =&gt; python
    [parent] =&gt; 
    [name] =&gt; Python
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; Python is a programming language that is heavily used in {{machine learning}}, in part because of the ectebsive libarraies that have been developed and in part because of the success of {{Jupyter notebook}}, which makes it easy to experiment, modify and share not just programs or the output of a session of actvity, but also the details of the process itself in a manner that can be replayed and modified by others.
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/aibook/quadtree">quadtree</a></span><!--  Array
(
    [slug] =&gt; quadtree
    [parent] =&gt; 
    [name] =&gt; quadtree
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; The quadtree is a tree-based data representation to store and index location information about a 2D image or area.  The root of the tree has four branches corresponding to the four quadrants of the image.  If there are no data points in any of these quadrants, then they are marked as empty.  Otherwise the points are allocated to the relevant quadrant and then the same process is used to build a quadtree of those.
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/aibook/quadstores">quadstores</a></span><!--  Array
(
    [slug] =&gt; quadstores
    [parent] =&gt; 
    [name] =&gt; quadstores
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; Quadstores are a form of {{database}} used to store {{semantic web}} data.. {{RDF}} is encoded as {{subject-predicate-object triples}} and often stored in a {{triplestore}}.  A quadstore allows several graphs of RDF data to be stored in teh same database.  Each item in the quadstores is of the form ( graph URI, subject, predicate, object ) and it is possible to query over a particular graph, all graphs or other combinations of graphs.  The graph URI can itself be used as the subject in assertions allowing {{provenance}} to be represented, for example, the origin, authority or relaibility of  a particualr sub-collection (graph) of triples.
)
  --></li>
  <li ><span class="name"><a href="https://alandix.com/glossary/aibook/qualia">qualia</a></span><!--  Array
(
    [slug] =&gt; qualia
    [parent] =&gt; 
    [name] =&gt; qualia
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; 
)
  --></li>
  <li ><span class="name"><a href="https://alandix.com/glossary/aibook/quantifiers">quantifiers</a></span><!--  Array
(
    [slug] =&gt; quantifiers
    [parent] =&gt; 
    [name] =&gt; quantifiers
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; 
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/aibook/quantum computing">quantum computing</a></span><!--  Array
(
    [slug] =&gt; quantum computing
    [parent] =&gt; 
    [name] =&gt; quantum computing
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; Quantum computers use various forms of quantum-level processs as an explicit part of their hardware.   They make extensive use of {[qubits}} physcial elements that are in multiple superimposed states, sort of 0 and 1 at the same time.  Quantum effects mean that effectively the same pieces of phsyical hardware simulrtaneosly perofrm multiple veesions of the same calculations, in contrats to classicla {{parallel processing}} where several physically dstinct processors do work on parts or variants of an algorith at the same time.  Ths may be used ot modle quantum-level effects in phsyics, but can also be used in other kinds of algoriths.  In particular many of the techniques used in {{cryptography}} will become insecure as a quantum computer can effectively try billions of options of a key similtaneously.  In AI, one can imagie that certain kinds of {{neural network}} could benefit and one of the technologies for {{reservoir computing}}  is quantum; however practical applications are probably still some years off  Note that conventional computers may experience quantum effects such as tunneling between adjacent tracks, but this is usually regarded as a problem (indeed is one of the limits of convetional electronics), whereas quantum computers make deliberate use of these kinds of effect.
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/aibook/quartile">quartile</a></span><!--  Array
(
    [slug] =&gt; quartile
    [parent] =&gt; 
    [name] =&gt; quartile
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; The quartiles of a {{probability distribution}} refer to the points (or soemtimes the intervals between them) that divide the paramter into four pieces where the probability of each is exactly a quarter.  This means that on average 25% of the time the value will lie below the {{lower quartile}} and 25% of the time larger than the {{upper quartile}}.  Arguably the {{median}} can be regarded as the middle quartile, but more often the term is sued only for the lowert and upper quartile.  
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/aibook/quasi-periodic">quasi-periodic</a></span><!--  Array
(
    [slug] =&gt; quasi-periodic
    [parent] =&gt; 
    [name] =&gt; quasi-periodic
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; A {{quasi-periodic}} phenomenon is one that almost has a periodic behaviour, but is not entirely predictable.  They may occur when otherwise periodic processes have noise.  For example, imagine a pedulum clock in a very windy location, the pendulum naturally swings once a secomd, but the gusts and breezes mean it may drift, with some swings a little faster and some a little slower.  Critically, if two such pendula were in different environments and started at the same moment, after a while the pendula would both still be approximately once a second, but may be out of {{phase}} with one another. 
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/aibook/query-by-browsing">Query-by-Browsing</a></span><!--  Array
(
    [slug] =&gt; query-by-browsing
    [parent] =&gt; 
    [name] =&gt; Query-by-Browsing
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; Query-by-Browsing (QbB) is an intelligent database query interface that allows users to choose records of interst from a listing and then infers a query, for example in {{SQL}}.  This is of course the opposite of most database query interfaces (including so called 'query by example') where the user writes.a query as input and then sees the results as output.
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/aibook/quinlan, ross">Quinlan, Ross</a></span><!--  Array
(
    [slug] =&gt; quinlan, ross
    [parent] =&gt; 
    [name] =&gt; Quinlan, Ross
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; Ross Quinlan developed the influential {{ID3}} and {{C4.5}} algorithms for inductive learning of {{decision trees}}
)
  --></li>
  <li ><span class="name"><a href="https://alandix.com/glossary/aibook/radial basis functions">radial basis functions</a></span><!--  Array
(
    [slug] =&gt; radial basis functions
    [parent] =&gt; 
    [name] =&gt; radial basis functions
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; 
)
  --></li>
  <li ><span class="name">random decision forest</span> - See <span class="has_content"><a href="https://alandix.com/glossary/aibook/random forest">random forest</a></span><!--  Array
(
    [slug] =&gt; random decision forest
    [parent] =&gt; 
    [name] =&gt; random decision forest
    [canonical] =&gt; 
    [see] =&gt; random forest
    [description] =&gt; 
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/aibook/random forest">random forest</a></span><!--  Array
(
    [slug] =&gt; random forest
    [parent] =&gt; 
    [name] =&gt; random forest
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; A random forest is a form of {{ensemble method}} for machine learning.  Large numbers of {{decision trees}} are generated by choosing random subsets of features from random subsets of the training data.  The indivdiual trees each give their own results and a separate layer of processing combines the results from all of the trees to give a final result.  While compartively simple, random forests have proved highly effective in a wide variey of application domains.
)
  --></li>
  <li ><span class="name"><a href="https://alandix.com/glossary/aibook/random walk">random walk</a></span><!--  Array
(
    [slug] =&gt; random walk
    [parent] =&gt; 
    [name] =&gt; random walk
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; 
)
  --></li>
  <li ><span class="name"><a href="https://alandix.com/glossary/aibook/randomising the order">randomising the order</a></span><!--  Array
(
    [slug] =&gt; randomising the order
    [parent] =&gt; 
    [name] =&gt; randomising the order
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; 
)
  --></li>
  <li ><span class="name">RBM</span> - See <span class="has_content"><a href="https://alandix.com/glossary/aibook/restricted boltzmann machine">Restricted Boltzmann Machine</a></span><!--  Array
(
    [slug] =&gt; rbm
    [parent] =&gt; 
    [name] =&gt; RBM
    [canonical] =&gt; 
    [see] =&gt; restricted boltzmann machine
    [description] =&gt; 
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/aibook/rdf">RDF</a></span><!--  Array
(
    [slug] =&gt; rdf
    [parent] =&gt; 
    [name] =&gt; RDF
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; RDF is the {{semantic web}} technology used to experess properties of web pages and in general {{URI}}s oncluding those representing things outside of the web.  RDF is expressed using {{subject-predicate-object}} tripples, where both the subject and predicate have to be {{URI}}s, but the object may also be soem form of literal value such as a number or string.  This allows one to express statements such as: &lt;br&gt; &amp;lt;https://alandix.com/academic/&amp;gt; &amp;lt;ex:is_about&amp;gt; &quot;Alan Dix&quot;
)
  --></li>
  <li ><span class="name"><a href="https://alandix.com/glossary/aibook/rdfs">RDFS</a></span><!--  Array
(
    [slug] =&gt; rdfs
    [parent] =&gt; 
    [name] =&gt; RDFS
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; 
)
  --></li>
  <li ><span class="name"><a href="https://alandix.com/glossary/aibook/reactive movement">reactive movement</a></span><!--  Array
(
    [slug] =&gt; reactive movement
    [parent] =&gt; 
    [name] =&gt; reactive movement
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; 
)
  --></li>
  <li ><span class="name"><a href="https://alandix.com/glossary/aibook/readout">readout</a></span><!--  Array
(
    [slug] =&gt; readout
    [parent] =&gt; 
    [name] =&gt; readout
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; 
)
  --></li>
  <li ><span class="name"><a href="https://alandix.com/glossary/aibook/reasoning">reasoning</a></span>- has subentries<ul class="subentries"><li ><span class="name">abductive</span> - See <span class="has_content"><a href="https://alandix.com/glossary/aibook/abductive reasoning">abductive reasoning</a></span><!--  Array
(
    [slug] =&gt; reasoning!abductive
    [parent] =&gt; reasoning
    [name] =&gt; abductive
    [canonical] =&gt; 
    [see] =&gt; abductive reasoning
    [description] =&gt; 
)
  --></li>
<li ><span class="name">analogy</span> - See <span class="has_content"><a href="https://alandix.com/glossary/aibook/reasoning by analogy">reasoning by analogy</a></span><!--  Array
(
    [slug] =&gt; reasoning!analogy
    [parent] =&gt; reasoning
    [name] =&gt; analogy
    [canonical] =&gt; 
    [see] =&gt; reasoning by analogy
    [description] =&gt; 
)
  --></li>
<li ><span class="name">antecedent-driven</span> - See <span class="has_content"><a href="https://alandix.com/glossary/aibook/backward reasoning">backward reasoning</a></span><!--  Array
(
    [slug] =&gt; reasoning!antecedent-driven
    [parent] =&gt; reasoning
    [name] =&gt; antecedent-driven
    [canonical] =&gt; 
    [see] =&gt; backward reasoning
    [description] =&gt; 
)
  --></li>
<li ><span class="name">backward</span> - See <span class="has_content"><a href="https://alandix.com/glossary/aibook/backward reasoning">backward reasoning</a></span><!--  Array
(
    [slug] =&gt; reasoning!backward
    [parent] =&gt; reasoning
    [name] =&gt; backward
    [canonical] =&gt; 
    [see] =&gt; backward reasoning
    [description] =&gt; 
)
  --></li>
<li ><span class="name">Bayesian</span> - See <span ><a href="https://alandix.com/glossary/aibook/bayesian reasoning">Bayesian reasoning</a></span><!--  Array
(
    [slug] =&gt; reasoning!bayesian
    [parent] =&gt; reasoning
    [name] =&gt; Bayesian
    [canonical] =&gt; 
    [see] =&gt; bayesian reasoning
    [description] =&gt; 
)
  --></li>
<li ><span class="name"><a href="https://alandix.com/glossary/aibook/reasoning!bottom-up">bottom-up</a></span><!--  Array
(
    [slug] =&gt; reasoning!bottom-up
    [parent] =&gt; reasoning
    [name] =&gt; bottom-up
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; 
)
  --></li>
<li ><span class="name">case-based</span> - See <span class="has_content"><a href="https://alandix.com/glossary/aibook/case-based reasoning">case-based reasoning</a></span><!--  Array
(
    [slug] =&gt; reasoning!case-based
    [parent] =&gt; reasoning
    [name] =&gt; case-based
    [canonical] =&gt; 
    [see] =&gt; case-based reasoning
    [description] =&gt; 
)
  --></li>
<li ><span class="name">certainty factors</span> - See <span class="has_content"><a href="https://alandix.com/glossary/aibook/certainty factors">certainty factors</a></span><!--  Array
(
    [slug] =&gt; reasoning!certainty factors
    [parent] =&gt; reasoning
    [name] =&gt; certainty factors
    [canonical] =&gt; 
    [see] =&gt; certainty factors
    [description] =&gt; 
)
  --></li>
<li ><span class="name">consequent-driven</span> - See <span class="has_content"><a href="https://alandix.com/glossary/aibook/backward reasoning">backward reasoning</a></span><!--  Array
(
    [slug] =&gt; reasoning!consequent-driven
    [parent] =&gt; reasoning
    [name] =&gt; consequent-driven
    [canonical] =&gt; 
    [see] =&gt; backward reasoning
    [description] =&gt; 
)
  --></li>
<li ><span class="name">data-driven</span> - See <span class="has_content"><a href="https://alandix.com/glossary/aibook/forward reasoning">forward reasoning</a></span><!--  Array
(
    [slug] =&gt; reasoning!data-driven
    [parent] =&gt; reasoning
    [name] =&gt; data-driven
    [canonical] =&gt; 
    [see] =&gt; forward reasoning
    [description] =&gt; 
)
  --></li>
<li ><span class="name">deductive</span> - See <span class="has_content"><a href="https://alandix.com/glossary/aibook/deductive reasoning">deductive reasoning</a></span><!--  Array
(
    [slug] =&gt; reasoning!deductive
    [parent] =&gt; reasoning
    [name] =&gt; deductive
    [canonical] =&gt; 
    [see] =&gt; deductive reasoning
    [description] =&gt; 
)
  --></li>
<li ><span class="name">forward</span> - See <span class="has_content"><a href="https://alandix.com/glossary/aibook/forward reasoning">forward reasoning</a></span><!--  Array
(
    [slug] =&gt; reasoning!forward
    [parent] =&gt; reasoning
    [name] =&gt; forward
    [canonical] =&gt; 
    [see] =&gt; forward reasoning
    [description] =&gt; 
)
  --></li>
<li ><span class="name">fuzzy</span> - See <span ><a href="https://alandix.com/glossary/aibook/fuzzy reasoning">fuzzy reasoning</a></span><!--  Array
(
    [slug] =&gt; reasoning!fuzzy
    [parent] =&gt; reasoning
    [name] =&gt; fuzzy
    [canonical] =&gt; 
    [see] =&gt; fuzzy reasoning
    [description] =&gt; 
)
  --></li>
<li ><span class="name">goal-driven</span> - See <span class="has_content"><a href="https://alandix.com/glossary/aibook/backward reasoning">backward reasoning</a></span><!--  Array
(
    [slug] =&gt; reasoning!goal-driven
    [parent] =&gt; reasoning
    [name] =&gt; goal-driven
    [canonical] =&gt; 
    [see] =&gt; backward reasoning
    [description] =&gt; 
)
  --></li>
<li ><span class="name">hypothesis-driven</span> - See <span ><a href="https://alandix.com/glossary/aibook/hypothesis-driven reasoning">hypothesis-driven reasoning</a></span><!--  Array
(
    [slug] =&gt; reasoning!hypothesis-driven
    [parent] =&gt; reasoning
    [name] =&gt; hypothesis-driven
    [canonical] =&gt; 
    [see] =&gt; hypothesis-driven reasoning
    [description] =&gt; 
)
  --></li>
<li ><span class="name">inductive</span> - See <span ><a href="https://alandix.com/glossary/aibook/inductive reasoning">inductive reasoning</a></span><!--  Array
(
    [slug] =&gt; reasoning!inductive
    [parent] =&gt; reasoning
    [name] =&gt; inductive
    [canonical] =&gt; 
    [see] =&gt; inductive reasoning
    [description] =&gt; 
)
  --></li>
<li ><span class="name">monotonic</span> - See <span ><a href="https://alandix.com/glossary/aibook/monotonic reasoning">monotonic reasoning</a></span><!--  Array
(
    [slug] =&gt; reasoning!monotonic
    [parent] =&gt; reasoning
    [name] =&gt; monotonic
    [canonical] =&gt; 
    [see] =&gt; monotonic reasoning
    [description] =&gt; 
)
  --></li>
<li ><span class="name">non-monotonic</span> - See <span class="has_content"><a href="https://alandix.com/glossary/aibook/non-monotonic reasoning">non-monotonic reasoning</a></span><!--  Array
(
    [slug] =&gt; reasoning!non-monotonic
    [parent] =&gt; reasoning
    [name] =&gt; non-monotonic
    [canonical] =&gt; 
    [see] =&gt; non-monotonic reasoning
    [description] =&gt; 
)
  --></li>
<li ><span class="name">probabilistic</span> - See <span ><a href="https://alandix.com/glossary/aibook/probabilistic reasoning">probabilistic reasoning</a></span><!--  Array
(
    [slug] =&gt; reasoning!probabilistic
    [parent] =&gt; reasoning
    [name] =&gt; probabilistic
    [canonical] =&gt; 
    [see] =&gt; probabilistic reasoning
    [description] =&gt; 
)
  --></li>
<li ><span class="name">top-down</span> - See <span ><a href="https://alandix.com/glossary/aibook/top-down reasoning">top-down reasoning</a></span><!--  Array
(
    [slug] =&gt; reasoning!top-down
    [parent] =&gt; reasoning
    [name] =&gt; top-down
    [canonical] =&gt; 
    [see] =&gt; top-down reasoning
    [description] =&gt; 
)
  --></li></ul><!--  Array
(
    [slug] =&gt; reasoning
    [parent] =&gt; 
    [name] =&gt; reasoning
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; 
    [subentries] =&gt; Array
        (
            [reasoning!abductive] =&gt; Array
                (
                    [slug] =&gt; reasoning!abductive
                    [parent] =&gt; reasoning
                    [name] =&gt; abductive
                    [canonical] =&gt; 
                    [see] =&gt; abductive reasoning
                    [description] =&gt; 
                )

            [reasoning!analogy] =&gt; Array
                (
                    [slug] =&gt; reasoning!analogy
                    [parent] =&gt; reasoning
                    [name] =&gt; analogy
                    [canonical] =&gt; 
                    [see] =&gt; reasoning by analogy
                    [description] =&gt; 
                )

            [reasoning!antecedent-driven] =&gt; Array
                (
                    [slug] =&gt; reasoning!antecedent-driven
                    [parent] =&gt; reasoning
                    [name] =&gt; antecedent-driven
                    [canonical] =&gt; 
                    [see] =&gt; backward reasoning
                    [description] =&gt; 
                )

            [reasoning!backward] =&gt; Array
                (
                    [slug] =&gt; reasoning!backward
                    [parent] =&gt; reasoning
                    [name] =&gt; backward
                    [canonical] =&gt; 
                    [see] =&gt; backward reasoning
                    [description] =&gt; 
                )

            [reasoning!bayesian] =&gt; Array
                (
                    [slug] =&gt; reasoning!bayesian
                    [parent] =&gt; reasoning
                    [name] =&gt; Bayesian
                    [canonical] =&gt; 
                    [see] =&gt; bayesian reasoning
                    [description] =&gt; 
                )

            [reasoning!bottom-up] =&gt; Array
                (
                    [slug] =&gt; reasoning!bottom-up
                    [parent] =&gt; reasoning
                    [name] =&gt; bottom-up
                    [canonical] =&gt; 
                    [see] =&gt; 
                    [description] =&gt; 
                )

            [reasoning!case-based] =&gt; Array
                (
                    [slug] =&gt; reasoning!case-based
                    [parent] =&gt; reasoning
                    [name] =&gt; case-based
                    [canonical] =&gt; 
                    [see] =&gt; case-based reasoning
                    [description] =&gt; 
                )

            [reasoning!certainty factors] =&gt; Array
                (
                    [slug] =&gt; reasoning!certainty factors
                    [parent] =&gt; reasoning
                    [name] =&gt; certainty factors
                    [canonical] =&gt; 
                    [see] =&gt; certainty factors
                    [description] =&gt; 
                )

            [reasoning!consequent-driven] =&gt; Array
                (
                    [slug] =&gt; reasoning!consequent-driven
                    [parent] =&gt; reasoning
                    [name] =&gt; consequent-driven
                    [canonical] =&gt; 
                    [see] =&gt; backward reasoning
                    [description] =&gt; 
                )

            [reasoning!data-driven] =&gt; Array
                (
                    [slug] =&gt; reasoning!data-driven
                    [parent] =&gt; reasoning
                    [name] =&gt; data-driven
                    [canonical] =&gt; 
                    [see] =&gt; forward reasoning
                    [description] =&gt; 
                )

            [reasoning!deductive] =&gt; Array
                (
                    [slug] =&gt; reasoning!deductive
                    [parent] =&gt; reasoning
                    [name] =&gt; deductive
                    [canonical] =&gt; 
                    [see] =&gt; deductive reasoning
                    [description] =&gt; 
                )

            [reasoning!forward] =&gt; Array
                (
                    [slug] =&gt; reasoning!forward
                    [parent] =&gt; reasoning
                    [name] =&gt; forward
                    [canonical] =&gt; 
                    [see] =&gt; forward reasoning
                    [description] =&gt; 
                )

            [reasoning!fuzzy] =&gt; Array
                (
                    [slug] =&gt; reasoning!fuzzy
                    [parent] =&gt; reasoning
                    [name] =&gt; fuzzy
                    [canonical] =&gt; 
                    [see] =&gt; fuzzy reasoning
                    [description] =&gt; 
                )

            [reasoning!goal-driven] =&gt; Array
                (
                    [slug] =&gt; reasoning!goal-driven
                    [parent] =&gt; reasoning
                    [name] =&gt; goal-driven
                    [canonical] =&gt; 
                    [see] =&gt; backward reasoning
                    [description] =&gt; 
                )

            [reasoning!hypothesis-driven] =&gt; Array
                (
                    [slug] =&gt; reasoning!hypothesis-driven
                    [parent] =&gt; reasoning
                    [name] =&gt; hypothesis-driven
                    [canonical] =&gt; 
                    [see] =&gt; hypothesis-driven reasoning
                    [description] =&gt; 
                )

            [reasoning!inductive] =&gt; Array
                (
                    [slug] =&gt; reasoning!inductive
                    [parent] =&gt; reasoning
                    [name] =&gt; inductive
                    [canonical] =&gt; 
                    [see] =&gt; inductive reasoning
                    [description] =&gt; 
                )

            [reasoning!monotonic] =&gt; Array
                (
                    [slug] =&gt; reasoning!monotonic
                    [parent] =&gt; reasoning
                    [name] =&gt; monotonic
                    [canonical] =&gt; 
                    [see] =&gt; monotonic reasoning
                    [description] =&gt; 
                )

            [reasoning!non-monotonic] =&gt; Array
                (
                    [slug] =&gt; reasoning!non-monotonic
                    [parent] =&gt; reasoning
                    [name] =&gt; non-monotonic
                    [canonical] =&gt; 
                    [see] =&gt; non-monotonic reasoning
                    [description] =&gt; 
                )

            [reasoning!probabilistic] =&gt; Array
                (
                    [slug] =&gt; reasoning!probabilistic
                    [parent] =&gt; reasoning
                    [name] =&gt; probabilistic
                    [canonical] =&gt; 
                    [see] =&gt; probabilistic reasoning
                    [description] =&gt; 
                )

            [reasoning!top-down] =&gt; Array
                (
                    [slug] =&gt; reasoning!top-down
                    [parent] =&gt; reasoning
                    [name] =&gt; top-down
                    [canonical] =&gt; 
                    [see] =&gt; top-down reasoning
                    [description] =&gt; 
                )

        )

)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/aibook/reasoning by analogy">reasoning by analogy</a></span><!--  Array
(
    [slug] =&gt; reasoning by analogy
    [parent] =&gt; 
    [name] =&gt; reasoning by analogy
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; Analogies are used extensively in {{human reasoning}} to help to understand new situations and to suggest potential problems solutions, for example, given the analogy &quot; a school teacher with their pupils are like a shepherd herding sheep&quot;, one might reason that as shepherds are better able to direct sheep is they are kept in a group, this might be a good strtegy for schoolteachers too.  Analogies can also be used by AI systems by noting structural similarities between past cases and the current situation.  As the analogy may fail to hold at some point, analogical reasoning may be used as a {{heuristic}}.  Within AI, the uses of AI are usually either as a {{transformational analogy}} (using at solutions to analogous problems) or a {{derivational analogy}} (using the process of obtaining the analogous solution).  Reasoning by analogy is often alongside {{case-based reasoning}}.
)
  --></li>
  <li ><span class="name"><a href="https://alandix.com/glossary/aibook/reasoning systems">reasoning systems</a></span><!--  Array
(
    [slug] =&gt; reasoning systems
    [parent] =&gt; 
    [name] =&gt; reasoning systems
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; 
)
  --></li>
  <li ><span class="name">reasoning with certainty factors</span> - See <span class="has_content"><a href="https://alandix.com/glossary/aibook/certainty factors">certainty factors</a></span><!--  Array
(
    [slug] =&gt; reasoning with certainty factors
    [parent] =&gt; 
    [name] =&gt; reasoning with certainty factors
    [canonical] =&gt; 
    [see] =&gt; certainty factors
    [description] =&gt; 
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/aibook/recall">recall</a></span><!--  Array
(
    [slug] =&gt; recall
    [parent] =&gt; 
    [name] =&gt; recall
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; When making a binary classification or decision, such as a disease diagnosis, recall is the {{probability}} that if the person really has the disease, or in general if the classification really is positive, then we  corectly make the decision, for example, if the person has the disease then we do diagnose it.   If a decision has a high recall, then the probability of a {{false negative}} is low.  Often increasing recall reduces {{precision}}; this is called the {{precision-recall trade-off}}.
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/aibook/recaptcha">reCAPTCHA</a></span><!--  Array
(
    [slug] =&gt; recaptcha
    [parent] =&gt; 
    [name] =&gt; reCAPTCHA
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; reCAPTCHA is a common security feature on web pages used to distinguish human users from automated {{bots}}.  reCAPTCHA appears as a small dialogue or popup on a website asking the user to perform a simple tasks such as read text or match images.  The tasks are designed to be simple for humans, but hard for machines, hence useful to distinguish then, indeed the term CAPTCHA is originally an acronym for &quot;Completely Automated Public {{Turing test}} to tell Computers and Humans Apart&quot;.  While this is their primary purpose, they are also used to  fill in gaps in AI algorithms and to create {{labelled training data}} for {{machine learning}}.  The original reCAPTCHA helped fill in difficult portions of OCR for scanned books and is an early example of {{human computation}}.
)
  --></li>
  <li ><span class="name">receiver operating characteristic</span> - See <span class="has_content"><a href="https://alandix.com/glossary/aibook/roc">ROC</a></span><!--  Array
(
    [slug] =&gt; receiver operating characteristic
    [parent] =&gt; 
    [name] =&gt; receiver operating characteristic
    [canonical] =&gt; 
    [see] =&gt; roc
    [description] =&gt; 
)
  --></li>
  <li ><span class="name"><a href="https://alandix.com/glossary/aibook/receptive field">receptive field</a></span><!--  Array
(
    [slug] =&gt; receptive field
    [parent] =&gt; 
    [name] =&gt; receptive field
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; 
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/aibook/recidivism">recidivism</a></span><!--  Array
(
    [slug] =&gt; recidivism
    [parent] =&gt; 
    [name] =&gt; recidivism
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; Recidivism is a technical term in the judicial system refering to the likelihood that soemone who has previsouly committed. acrime will do so again.  The controversial AI system {{COMPAS}} creates a precdicted recidivism score (effectively high or low likelhood of reoffending) and has been used by judges in the USA as part of parole decisions.  {{COMPAS}} was built using {{machine learning}} over past felons and their actual reoffending rates, so in a way is based on {{ground truth}} data.  However, in practice the scores appear to exhibit significant {{racial bias}}, made worse because {{automation bias}} means the scores may effectively be treated as recommendations.
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/aibook/recommender systems">recommender systems</a></span><!--  Array
(
    [slug] =&gt; recommender systems
    [parent] =&gt; 
    [name] =&gt; recommender systems
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; Recommender systems use data collected about past behaviours, such as books or music choices, of large numbers of people and use this to suggest new things to read, or listen to.  They use {{bog data}} echnques such as lareg-scale {{statistical methods}}.
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/aibook/recurrent neural network">recurrent neural network</a></span><!--  Array
(
    [slug] =&gt; recurrent neural network
    [parent] =&gt; 
    [name] =&gt; recurrent neural network
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; A recurrent neural network is a form of {{neural network}} with a state/memory that runs on {{time series}} or {{sequential}} data.  At each step a {{neura networ}} uses the current state and inout token to geerate the next state and to predict the next token.  As the hidden state is not part of the training data, the {{training phase}} is more complex than for a simple {{neural network}}.
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/aibook/reduce">reduce</a></span><!--  Array
(
    [slug] =&gt; reduce
    [parent] =&gt; 
    [name] =&gt; reduce
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; Reduce is a standard function in {{Lisp}} and other {{functional programming}} languages.  The basic reduce function has two arguments reduce(f,list), where the function f takes two arguments and is initially applied to the first two elements in the list, then the fucntion is applied again to the result and the third item, and so on. For example: &lt;br&gt; &amp;nbsp; &amp;nbsp; &amp;nbsp; reduce( '+', (1,2,3,4) )  =  (((1+2)+3)+4). &lt;br&gt; Reduce functions typically allow an optional third argument that is used as an initial value. &lt;br&gt;&lt;br&gt; The reduce function is one of the inspirations for the {{MapReduce}} infrastructure for massive scale processing of {{big data}}.
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/aibook/redundancy">redundancy</a></span><!--  Array
(
    [slug] =&gt; redundancy
    [parent] =&gt; 
    [name] =&gt; redundancy
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; Redundancy is when sensor readings or items of data replicate in full or part data that is sensed or stored elsewhere.  Redundancy can be seen as a source of {{inefficiency}} so that one may try to remove redundant values as part of {{data clearening}}.  However, redundancy can also be critical in {{fault tolerant}} systems allowinga. level of recovery if a sensor fails or data is lost.  Redundancy may be precise, in that one peoce of data can be calculated from others.  Alterbatively, the redundant data items may be statistically related, for example where several values in some way measure or represnt the same thing, but have a degree of noise.  In  this case {{statistical}} techniques, including simple averaging or {[time series}} techniques can be used to obtian a smaller number of more accurate readings.  In {{actvity recognition}} systems redundant sensors give more {[robust}} detection, but also mean that as new sensors are addedd, they can be {{correlated}} with existing sensors, adding to redundacy especially if existing sensors fail or are removed and allowing deployments of sensors to change whilst retaining higher-level reasoning.
)
  --></li>
  <li ><span class="name"><a href="https://alandix.com/glossary/aibook/region detection">region detection</a></span><!--  Array
(
    [slug] =&gt; region detection
    [parent] =&gt; 
    [name] =&gt; region detection
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; 
)
  --></li>
  <li ><span class="name"><a href="https://alandix.com/glossary/aibook/region growing">region growing</a></span><!--  Array
(
    [slug] =&gt; region growing
    [parent] =&gt; 
    [name] =&gt; region growing
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; 
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/aibook/regression line">regression line</a></span><!--  Array
(
    [slug] =&gt; regression line
    [parent] =&gt; 
    [name] =&gt; regression line
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; In {{linear regression}}, the {{regression line}} is the line that best fits the data.
)
  --></li>
  <li ><span class="name"><a href="https://alandix.com/glossary/aibook/regret">regret</a></span><!--  Array
(
    [slug] =&gt; regret
    [parent] =&gt; 
    [name] =&gt; regret
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; 
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/aibook/reinforcement">reinforcement</a></span><!--  Array
(
    [slug] =&gt; reinforcement
    [parent] =&gt; 
    [name] =&gt; reinforcement
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; Reinforcement is the process of increasing or decreasing the weightings or parameters that gave rise to a positive or negative  behaviour.  The idea is based on Skinner's experiments with pigeons in the 1950s.  If the mapoing between stimulus, behaviour and reward is simple and direct, for example a {{look-up table}} with weights, then reinforcement is relatuveky straughtfirward.  More commonly however, the {{causal relatiinship}} is complex and there are problems of {{credit assignment}}.
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/aibook/reinforcement learning">reinforcement learning</a></span><!--  Array
(
    [slug] =&gt; reinforcement learning
    [parent] =&gt; 
    [name] =&gt; reinforcement learning
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; Reinforcement learning is used to learn complex time and situation behaviours, when there is eiether no training data or it is insufficient.  Often the poitn ar which rewards occur are well after the action that caused them and may be the result of several past actions leading toe {{credit assignment}} problems, which make positive or negative {{reinforcement}} hard.  Learning takes place due to interactions with a (real ir simukated) world and therfore have a cost both directly due to the action being perfroemd (energy expenditure ofr a robot, network costs for a web agent) and indirecly due to the posutuve or negative effects of the action. However, without taking actions there is no potential for learning, this leads to an {{exploration-exploitation trade-off}}.
)
  --></li>
  <li ><span class="name"><a href="https://alandix.com/glossary/aibook/relational database">relational database</a></span><!--  Array
(
    [slug] =&gt; relational database
    [parent] =&gt; 
    [name] =&gt; relational database
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; 
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/aibook/relaxation term">relaxation term</a></span><!--  Array
(
    [slug] =&gt; relaxation term
    [parent] =&gt; 
    [name] =&gt; relaxation term
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; A relaxation term is used to govern the speen of adadptation of many kinds of {{iterative algorithm}}.  These algorithms are often have the form: &lt;br&gt; &amp;nbsp; &amp;nbsp; &amp;nbsp; new_value = old_value + &amp;lambda; change_direction. &lt;br&gt; The term &amp;lambda; is the relaxation term.  If this is too large the algorithm is {{unstable}} swinging wildly between different values, if it is too small the algorithm takes a very ling time to {{converge}}.  In some algoruthms, notably {{simulated annealing}} the relaxation terms starts relatively large (but not so large to cause instability) to allow rapid movement towards the general site of a solutuon, and then gradually reduced to allow fine tuning.
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/aibook/relevance feedback">relevance feedback</a></span><!--  Array
(
    [slug] =&gt; relevance feedback
    [parent] =&gt; 
    [name] =&gt; relevance feedback
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; Relevance feedback is used in {{docuemnt retreival}}, {{recommender systes}} and other forms of {{intelligent user interfaces}} to allow users to say how good autimated suggestions are.  This is then fed back into the algorithms to improve future suggestions.  In some cases relevance feedback is {[explicit}}, for example, adding a star rating or thumbs up, in others it is {{implicit}} for example, if the users clicks through for more information about a suggested item.  {{Epistemic interaction}} can be used to design interactions that increase this feedback.
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/aibook/relevance information">relevance information</a></span><!--  Array
(
    [slug] =&gt; relevance information
    [parent] =&gt; 
    [name] =&gt; relevance information
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; {{Web search}} systems may use various kinds of relevance information as well as the presence of gthe search terms in text.  For example, if search words occur close to one another, or if page linking to a specific page use the terms as their link text.
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/aibook/reparameterisation">reparameterisation</a></span><!--  Array
(
    [slug] =&gt; reparameterisation
    [parent] =&gt; 
    [name] =&gt; reparameterisation
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; Reparamerisation is where one transforms the parameters that describe a problem or solution space in such a way that preserves information, but may be easier to process or learn from.  For example, changing a date represented as DD/MM/YYY into seconds since 1970 in order to facilitate rapid comparisons.
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/aibook/res cogitans">res cogitans</a></span><!--  Array
(
    [slug] =&gt; res cogitans
    [parent] =&gt; 
    [name] =&gt; res cogitans
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; Res cogitans is the term used by {{Descartes}} to refer to mental stuff, which he regarded as different from {{res extensa}} the physical material of our bodies.  This is a form of {{dualism}}, which is rejected by most modern philosophers ... except insofar as many feel that biological brains are intrinsically different from computers.
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/aibook/res extensa">res extensa</a></span><!--  Array
(
    [slug] =&gt; res extensa
    [parent] =&gt; 
    [name] =&gt; res extensa
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; Res extensa is the term used by {{Descartes}} to refer to the physical material of our bodies, which he regarded as different from {{res cogitans}} the mental stuff that allows us to think.  This is a form of {{dualism}}, which is rejected by most modern philosophers ... except insofar as many feel that biological brains are intrinsically different from computers.
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/aibook/rescue dogs">rescue dogs</a></span><!--  Array
(
    [slug] =&gt; rescue dogs
    [parent] =&gt; 
    [name] =&gt; rescue dogs
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; Within the book 'rescue dogs' refers to search and rescue, for example the St Bernard used in mountain rescue.  However, the term can also mean dogs that are waiting to be re-homed.
)
  --></li>
  <li ><span class="name"><a href="https://alandix.com/glossary/aibook/responsibility">responsibility</a></span><!--  Array
(
    [slug] =&gt; responsibility
    [parent] =&gt; 
    [name] =&gt; responsibility
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; 
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/aibook/reservoir computing">reservoir computing</a></span><!--  Array
(
    [slug] =&gt; reservoir computing
    [parent] =&gt; 
    [name] =&gt; reservoir computing
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; Reservoir computing uses a variety of physical, biologcal or computational processes to increase the {{non-linear diversity}} of input data the outputs of whih can then be used as inputs for a simpler final {{machine learning}} stage..  For example, input data might be used to drive electrical impolses into a semi-chaotic silicon substrate and the output currents measured at multiple points.
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/aibook/residuals">residuals</a></span><!--  Array
(
    [slug] =&gt; residuals
    [parent] =&gt; 
    [name] =&gt; residuals
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; The residuals are the difference between actual data values and their modelled values after some sort of {{data fitting}} process.  For example in {{linear regression}} the distance of the data points from the {{regression line}}.  In general, for data of the form (x&lt;sub&gt;i&lt;/sub&gt;,y(sub&gt;i&lt;/sub&gt;) if a model 'M' has been fitted to predict y form x, then the residuals r&lt;sub&gt;i&lt;/sub&gt; are given by: &lt;br&gt; &amp;nbsp; &amp;nbsp; &amp;nbsp; r&lt;sub&gt;i&lt;/sub&gt; &amp;nbsp; = &amp;nbsp; y(sub&gt;i&lt;/sub&gt; &ndash; M(x&lt;sub&gt;i&lt;/sub&gt;) &lt;br&gt; Often the model fitting process is expressed in reducing the {[sum of squares}} of the residuals or some similar metric as a {{fitness function}}.
)
  --></li>
  <li ><span class="name">resource description framework</span> - See <span class="has_content"><a href="https://alandix.com/glossary/aibook/rdf">RDF</a></span><!--  Array
(
    [slug] =&gt; resource description framework
    [parent] =&gt; 
    [name] =&gt; resource description framework
    [canonical] =&gt; 
    [see] =&gt; rdf
    [description] =&gt; 
)
  --></li>
  <li ><span class="name">resource description framework schema</span> - See <span ><a href="https://alandix.com/glossary/aibook/rdfs">RDFS</a></span><!--  Array
(
    [slug] =&gt; resource description framework schema
    [parent] =&gt; 
    [name] =&gt; resource description framework schema
    [canonical] =&gt; 
    [see] =&gt; rdfs
    [description] =&gt; 
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/aibook/responsible innovation">responsible innovation</a></span><!--  Array
(
    [slug] =&gt; responsible innovation
    [parent] =&gt; 
    [name] =&gt; responsible innovation
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; Responsible innovaton (RI) means understanding the potential ethical implications of one's work including both societal and environmental impacts and both the outcomes and the process by which it is performed.  For example, you may be developing a new system for a charity (ethically good outcome), but not think about the impact on the participants of the initial prototype when it is closed down.  The impacts may also be indirect or future for example, developing systems to improve urban traffic flow might have short term postive impacts, but lead to more car use in the longer term.  These issues apply both the product deelopment and reesearch.  For the latter one often has to imagine the longer term trajectory of the research -- the UK research funding agency EPSRC have developed the {{AREA framework}} to help researchers in the RI process.
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/aibook/restricted boltzmann machine">Restricted Boltzmann Machine</a></span><!--  Array
(
    [slug] =&gt; restricted boltzmann machine
    [parent] =&gt; 
    [name] =&gt; Restricted Boltzmann Machine
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; A Restricted Boltzmann Machine (RBM) is a form of {{Boltzmann Machine}} where input and output layers are fully connected to each another, but have no internal connections.  That is gthere are no input-input, not output-output links.  Otherwsie they are like a standard {{Boltzmann Machine} } with a training phase with inputs clamped and outputs free to vary and then a recall mode where some inputs are clamped and the network fills in the gaps.  As well as this reocnstreuction mode, RBMs are often used for {{feature detction}} in the early stages of {{deep neaural networks}} as they use feedforward learning rather than {{backpropagation}}.      The Restricted Boltzmann Machine (RBM) like the {{Boltzmann Machine}} is a form of {{neural network}} and {{autoencoder}}.
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/aibook/ridge">ridge</a></span><!--  Array
(
    [slug] =&gt; ridge
    [parent] =&gt; 
    [name] =&gt; ridge
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; A ridge in the {{fitness landscape}} is when there is a direction of change in the parameters that makes very little difference to the {{fitness function}}, but where small changes in any other direction drop the fitness radically.  For example, suppose the fitness in an area is approximately 1&amp;minus;&amp;radic;| x&amp;minus;y | and we start at the (x,y) point (2,2) where the fitness function is zero.  Small changes to (2+&amp;delta;&lt;sub&gt;x&lt;/sub&gt;, 2+&amp;delta;&lt;sub&gt;y&lt;/sub&gt;) stay at one if &amp;delta;&lt;sub&gt;x&lt;/sub&gt;=&amp;delta;&lt;sub&gt;y&lt;/sub&gt;, but if they are not equal even quite small changes where &amp;delta;&lt;sub&gt;x&lt;/sub&gt;&amp;ne;&amp;delta;&lt;sub&gt;y&lt;/sub&gt; will lead to rapid drops in the fitness function;
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/aibook/risk avoidance">risk avoidance</a></span><!--  Array
(
    [slug] =&gt; risk avoidance
    [parent] =&gt; 
    [name] =&gt; risk avoidance
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; Risk avoidance is where one deiberately adopts a strtegy hat reduces {{risk}}, for example prefering solutions in a {{gam space}} that have lower expected payback, but less variability.  For exampel, the {{minimax algotihm}} is effectively about minimising the worst case loss against a perfect player.  A less {{risk averse}} approach against an imperfect opponent might be to choose a path that has a small potential to lead to a bad defeat, but has also many ways once can turn the game around and win.  {{Stochastic games}} where there is an element of chance may mean an AI player has to choose between options that have low potential yield, but also low {{uncertainty}}, and those with higher expected gains,but with a lot of {{uncertainty}}.  By  adjusting the balance between these one can make the asystem more or less {{risk averse}}. 
)
  --></li>
  <li ><span class="name">RNN</span> - See <span class="has_content"><a href="https://alandix.com/glossary/aibook/recurrent neural network">recurrent neural network</a></span><!--  Array
(
    [slug] =&gt; rnn
    [parent] =&gt; 
    [name] =&gt; RNN
    [canonical] =&gt; 
    [see] =&gt; recurrent neural network
    [description] =&gt; 
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/aibook/robert wadlow">Robert Wadlow</a></span><!--  Array
(
    [slug] =&gt; robert wadlow
    [parent] =&gt; 
    [name] =&gt; Robert Wadlow
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; Robert Wadlow was the tallest person ever recorded, standing 2.72 metres tall (8ft 11inches).  He is an example of an {{outlier}} that is real, albeit unusual, data rather than an error in data collection or processing.
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/aibook/robert's operator">Robert's operator</a></span><!--  Array
(
    [slug] =&gt; robert's operator
    [parent] =&gt; 
    [name] =&gt; Robert's operator
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; Robert's operator is a {{non-linear}} filter to detect the presence of lines in an image. It work son a ecah 2x2 portion of the scence and is defined as the sum of the absolute differences between opposing corners: &lt;br&gt; &amp;nbsp; &amp;nbsp; &amp;nbsp;  G(x,y)  &amp;nbsp;  =  &amp;nbsp;  | f(x,y) &amp;minus; f(x+1,y-1)  | &amp;nbsp; + &amp;nbsp;  | f(x+1,y) &amp;minus;  f(x,y-1) |$
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/aibook/robotics">robotics</a></span>- has subentries<ul class="subentries"><li ><span class="name">current state representation</span> - See <span class="has_content"><a href="https://alandix.com/glossary/aibook/current state representation">current state representation</a></span><!--  Array
(
    [slug] =&gt; robotics!current state representation
    [parent] =&gt; robotics
    [name] =&gt; current state representation
    [canonical] =&gt; 
    [see] =&gt; current state representation
    [description] =&gt; 
)
  --></li>
<li ><span class="name"><a href="https://alandix.com/glossary/aibook/robotics!feedback">feedback</a></span><!--  Array
(
    [slug] =&gt; robotics!feedback
    [parent] =&gt; robotics
    [name] =&gt; feedback
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; 
)
  --></li>
<li ><span class="name">historical representation</span> - See <span class="has_content"><a href="https://alandix.com/glossary/aibook/historical representation">historical representation</a></span><!--  Array
(
    [slug] =&gt; robotics!historical representation
    [parent] =&gt; robotics
    [name] =&gt; historical representation
    [canonical] =&gt; 
    [see] =&gt; historical representation
    [description] =&gt; 
)
  --></li>
<li ><span class="name"><a href="https://alandix.com/glossary/aibook/robotics!practical">practical</a></span><!--  Array
(
    [slug] =&gt; robotics!practical
    [parent] =&gt; robotics
    [name] =&gt; practical
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; 
)
  --></li>
<li ><span class="name"><a href="https://alandix.com/glossary/aibook/robotics!safety">safety</a></span><!--  Array
(
    [slug] =&gt; robotics!safety
    [parent] =&gt; robotics
    [name] =&gt; safety
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; 
)
  --></li></ul><!--  Array
(
    [slug] =&gt; robotics
    [parent] =&gt; 
    [name] =&gt; robotics
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; Robotics is the arae of AI and engineering that studies the physical construction and computational control of {{robots}}.
    [subentries] =&gt; Array
        (
            [robotics!current state representation] =&gt; Array
                (
                    [slug] =&gt; robotics!current state representation
                    [parent] =&gt; robotics
                    [name] =&gt; current state representation
                    [canonical] =&gt; 
                    [see] =&gt; current state representation
                    [description] =&gt; 
                )

            [robotics!feedback] =&gt; Array
                (
                    [slug] =&gt; robotics!feedback
                    [parent] =&gt; robotics
                    [name] =&gt; feedback
                    [canonical] =&gt; 
                    [see] =&gt; 
                    [description] =&gt; 
                )

            [robotics!historical representation] =&gt; Array
                (
                    [slug] =&gt; robotics!historical representation
                    [parent] =&gt; robotics
                    [name] =&gt; historical representation
                    [canonical] =&gt; 
                    [see] =&gt; historical representation
                    [description] =&gt; 
                )

            [robotics!practical] =&gt; Array
                (
                    [slug] =&gt; robotics!practical
                    [parent] =&gt; robotics
                    [name] =&gt; practical
                    [canonical] =&gt; 
                    [see] =&gt; 
                    [description] =&gt; 
                )

            [robotics!safety] =&gt; Array
                (
                    [slug] =&gt; robotics!safety
                    [parent] =&gt; robotics
                    [name] =&gt; safety
                    [canonical] =&gt; 
                    [see] =&gt; 
                    [description] =&gt; 
                )

        )

)
  --></li>
  <li ><span class="name"><a href="https://alandix.com/glossary/aibook/robots">robots</a></span><!--  Array
(
    [slug] =&gt; robots
    [parent] =&gt; 
    [name] =&gt; robots
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; 
)
  --></li>
  <li ><span class="name"><a href="https://alandix.com/glossary/aibook/robust to failure">robust to failure</a></span><!--  Array
(
    [slug] =&gt; robust to failure
    [parent] =&gt; 
    [name] =&gt; robust to failure
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; 
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/aibook/roc">ROC</a></span><!--  Array
(
    [slug] =&gt; roc
    [parent] =&gt; 
    [name] =&gt; ROC
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; A ROC diagram or ROC curve is a visualisation of the {{precision-recall trade-off}}.    A ROC diagram can also be used to help choise between classifiers. 
)
  --></li>
  <li ><span class="name">Ross Quinlan</span> - See <span class="has_content"><a href="https://alandix.com/glossary/aibook/quinlan, ross">Quinlan, Ross</a></span><!--  Array
(
    [slug] =&gt; ross quinlan
    [parent] =&gt; 
    [name] =&gt; Ross Quinlan
    [canonical] =&gt; 
    [see] =&gt; quinlan, ross
    [description] =&gt; 
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/aibook/rote learning">rote learning</a></span><!--  Array
(
    [slug] =&gt; rote learning
    [parent] =&gt; 
    [name] =&gt; rote learning
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; In rote learning one learns precisely the examples given.  For humans this can be hard, but for a machine it is easy, just store all training examples with ther exepcted output and then lookup the correct value.   On its own this cannot {{generalize}}, when faced with a novel input that it has not seen before it has no way of filling in or estimating the correct output based on the inputs.  While this is a very simple form of {{machine learning}}, it can be a valuable part of more complex learning systems.
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/aibook/route planning">route planning</a></span><!--  Array
(
    [slug] =&gt; route planning
    [parent] =&gt; 
    [name] =&gt; route planning
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; Route planning is about finding paths form one location to another usually restructed to particular kinds of connection (such as roads or railways) and often with some form of {{optimisation criterion}} (such as minimal fuel costs) and {{constraints}} (such as passing particular intermediate points, or avoiding narrow roads).  If all of the distances, links and locatiosn are known then various forms of graph search can be used, notably the {{A* algorithm}}.  Where some parts are not known in advance there may need to be some form of high-level global planning to determine a best-guess approxiomate route, with local planning if problems are encountered, such as.a blocked road.
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/aibook/rubin's vase">Rubin's vase</a></span><!--  Array
(
    [slug] =&gt; rubin's vase
    [parent] =&gt; 
    [name] =&gt; Rubin's vase
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; Rubin's vase is an {{ambiguous image}}, which can be seen as vase or as two faces looking at one another.  Human viewers tend to see such images flip {{quasi-periodically}} between the two interpretations: vase, then faces, then vase again ... This gives clues as to how the brain deals with ambiguity and hence lessons for computer {{disambiguation}}. 
)
  --></li>
  <li ><span class="name"><a href="https://alandix.com/glossary/aibook/rule induction">rule induction</a></span><!--  Array
(
    [slug] =&gt; rule induction
    [parent] =&gt; 
    [name] =&gt; rule induction
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; 
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/aibook/saccades">saccades</a></span><!--  Array
(
    [slug] =&gt; saccades
    [parent] =&gt; 
    [name] =&gt; saccades
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; Saccades are the raoid movements your eye makes over a scene as you take in the details.  They are needed because a small portion at the center of the retina, called the {{fovea}}, contans the majority of your light senisng cells.    The saccades allows you to build up an overall detailed view rather like taking lots of photographs and then stitching them together to form a panorama. 
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/aibook/sail databank">SAIL databank</a></span><!--  Array
(
    [slug] =&gt; sail databank
    [parent] =&gt; 
    [name] =&gt; SAIL databank
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; The SAIL databank is a large collection of interlinked datasest related to health data promarily in Wales.  It is used for social and health related research..   A central aspect of the datset is that it connects all the relevent data about an indvidual including hosipital visits, GP appointemnts and also data from local government etc.  In order to link new datsets all ingested data is matcehd against a special database of personally identifiable information, and then {{pseudonomised}} so that the unique id for each individual can be matched, but no identifiable information is present in the data saved for future processing.  Even given this {{security}} and {{privacy}} are major issues with stringent access controls and no direct internet connections.
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/aibook/salience">salience</a></span><!--  Array
(
    [slug] =&gt; salience
    [parent] =&gt; 
    [name] =&gt; salience
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; Salience is the extent to which an event, object, or other stimulus captures {{attention}}.  Typically things have high salience if they (i) stand out visually (or on other senory ways) from their backgorund; (ii) if they are less frequently experienced; or (ii) if they have greater potential impact.  For example, (i) a bright red flower in the middle of a green field; (ii) an ice cream in the middle of winter; or (iii) a lion prowling.  In fact a lion prowling in a shopping street is an example of both (ii) and (iii)!  Salince is crucial to help humans and other animals choose were to focus attention and hence ac and learn more effectively.  Computational models of {{attention}} have been particularly important in the development of {{large language models}}.
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/aibook/sampling bias">sampling bias</a></span><!--  Array
(
    [slug] =&gt; sampling bias
    [parent] =&gt; 
    [name] =&gt; sampling bias
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; Sampling bias occurs when the composition of your sample in some ways skews your results.  For example, if a market researcher goes into a shopping street on a weekday afternoon they will encounter few employed people.  
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/aibook/sanity checks">sanity checks</a></span><!--  Array
(
    [slug] =&gt; sanity checks
    [parent] =&gt; 
    [name] =&gt; sanity checks
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; Raw data may have errors, mistypings or other problems.  Sanity checks are small, usualy realtively simple, pieces of code run across a dataset to verify simple properties.  For example whether items indexed in a book all appear in the glossary, or whether a field that is intended to be a number indeed only contains numbers.  This is often a critical part of {{data cleaning}} (or {{data wrangling}})   In addition, when dealing with large quantities of data it can be hard to be sure that transformations, which appear sensible in code, actually do what was intended as only a small portion of the data can be inspected by hand.   Sanity checks can again be applied after a transformation stage to help build confidence.  Sanity checs often involve simple {{regular expression}} matching (e.g. to verify the number field) or looking up values (e.g. to check glossary items are defined).
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/aibook/satisficing">satisficing</a></span><!--  Array
(
    [slug] =&gt; satisficing
    [parent] =&gt; 
    [name] =&gt; satisficing
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; Satisficing is about acheiving good enough solutions, rtaher than necessarily striving for the very best.  The term was coined by {{Herbert A. Simon}}, the poliical economist and AI pioneer, as a blend between he satisfy and suffice.  {{Human reasoning}} is often driven by satisficing and {{bounded rationality}} as real world decisions often have the made in the absnce of complete information and without time or mental capacity to fully reason through all scenarios.  Although the memory cacpity of machines can allow greater amonts of lookahead, AI systems are ultimately faced with similar situations and this is one of the reasons for the use of {{heuristics}}.
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/aibook/scale-related variability">scale-related variability</a></span><!--  Array
(
    [slug] =&gt; scale-related variability
    [parent] =&gt; 
    [name] =&gt; scale-related variability
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; Scale-related variability is when the variablity, often measured by {{standard deviation}} or {{variance}}) is larger (or smaller) when the underlying value is larger.  For example, if you are measuring times, such the evaporation time of different liquids, it is likely that the variablity of those with long evaporation times is greater than those with fast evaporation.  Often {{log transforms} are used with time-measurements such as these to make the variance more uniform.  This is a form of {{pre-whitening}}.
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/aibook/schema.org">schema.org</a></span><!--  Array
(
    [slug] =&gt; schema.org
    [parent] =&gt; 
    [name] =&gt; schema.org
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; Schema.org is a set of standards for embadding semntic information in human-readable web pages. Various mechanisms had emerged as de-facto standards, including {{microformats}}, but aregly for very particular data types and applications.  Several major vendors worked together to create the standards in schema.org, partly to enable better {{web scraping}} by theor own {{crawlers}}.  Scema.org defines {{vocabularies}} of semantic terms (e.g. around personal information, or events) and multiple ways that these can be embedded into web pages {{backwards compatible}} with the earlier emergent standrads. 
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/aibook/scientific data">scientific data</a></span><!--  Array
(
    [slug] =&gt; scientific data
    [parent] =&gt; 
    [name] =&gt; scientific data
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; One application of AI, and in particular {{machine learning}}, is to process and interpret scientific data.  This can often be voluminous, indeed the {{world wide web}} was originally developed to help share the vast amunts of data generated at {{CERN}}, it is thus an early example of {{big data}}.  Some forms of processing may simply be to reduce the dataset, ignoring common events or patterns and bringing more interesting ones to the scientists' attention.  However, forms of {{rule induction}} can be used to infer scientific laws directly from data, in a similar way to how Keplar spoted the relationships between planets' orbital distance and speed, namely that they &quot;sweep out equal areas in equal times&quot;.  Automated systems are able to try many different forms of equations and then choose ones that are both simple and fit the data well.
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/aibook/screen scraping">screen scraping</a></span><!--  Array
(
    [slug] =&gt; screen scraping
    [parent] =&gt; 
    [name] =&gt; screen scraping
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; Screen scraping is the name given to various technologies, tools and pratical techniques that are used to extract information from comouter systems with no built-in data interface or {{API}}.  It is most commoinly applied to old 80x25 terminal-based screen interfaces, but can also be used (with more difficulty) for window-based applications.  Screen scraping can be used purely to obtain information from the current screen, but typically there will be techniques to drive the application, for example emulating key presses or muse clicks.  This can be use to {{crawl}} the application and extract data, or to allow the creation of an alternative (more up-to-date) {{user interface}}.  When screen scraping is applied to web pages it is called {{web scrapping}}.
)
  --></li>
  <li ><span class="name"><a href="https://alandix.com/glossary/aibook/script">script</a></span><!--  Array
(
    [slug] =&gt; script
    [parent] =&gt; 
    [name] =&gt; script
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; 
)
  --></li>
  <li ><span class="name"><a href="https://alandix.com/glossary/aibook/scrutability">scrutability</a></span><!--  Array
(
    [slug] =&gt; scrutability
    [parent] =&gt; 
    [name] =&gt; scrutability
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; 
)
  --></li>
  <li ><span class="name"><a href="https://alandix.com/glossary/aibook/search">search</a></span>- has subentries<ul class="subentries"><li ><span class="name"><a href="https://alandix.com/glossary/aibook/search!adversarial">adversarial</a></span><!--  Array
(
    [slug] =&gt; search!adversarial
    [parent] =&gt; search
    [name] =&gt; adversarial
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; 
)
  --></li>
<li ><span class="name">best first</span> - See <span class="has_content"><a href="https://alandix.com/glossary/aibook/best first search">best first search</a></span><!--  Array
(
    [slug] =&gt; search!best first
    [parent] =&gt; search
    [name] =&gt; best first
    [canonical] =&gt; 
    [see] =&gt; best first search
    [description] =&gt; 
)
  --></li>
<li ><span class="name">breadth first</span> - See <span class="has_content"><a href="https://alandix.com/glossary/aibook/breadth first search">breadth first search</a></span><!--  Array
(
    [slug] =&gt; search!breadth first
    [parent] =&gt; search
    [name] =&gt; breadth first
    [canonical] =&gt; 
    [see] =&gt; breadth first search
    [description] =&gt; 
)
  --></li>
<li ><span class="name">closed list</span> - See <span class="has_content"><a href="https://alandix.com/glossary/aibook/closed list">closed list</a></span><!--  Array
(
    [slug] =&gt; search!closed list
    [parent] =&gt; search
    [name] =&gt; closed list
    [canonical] =&gt; 
    [see] =&gt; closed list
    [description] =&gt; 
)
  --></li>
<li ><span class="name">depth first</span> - See <span ><a href="https://alandix.com/glossary/aibook/depth first search">depth first search</a></span><!--  Array
(
    [slug] =&gt; search!depth first
    [parent] =&gt; search
    [name] =&gt; depth first
    [canonical] =&gt; 
    [see] =&gt; depth first search
    [description] =&gt; 
)
  --></li>
<li ><span class="name">deterministic</span> - See <span class="has_content"><a href="https://alandix.com/glossary/aibook/deterministic search">deterministic search</a></span><!--  Array
(
    [slug] =&gt; search!deterministic
    [parent] =&gt; search
    [name] =&gt; deterministic
    [canonical] =&gt; 
    [see] =&gt; deterministic search
    [description] =&gt; 
)
  --></li>
<li ><span class="name">feasibility</span> - See <span ><a href="https://alandix.com/glossary/aibook/feasibility">feasibility</a></span><!--  Array
(
    [slug] =&gt; search!feasibility
    [parent] =&gt; search
    [name] =&gt; feasibility
    [canonical] =&gt; 
    [see] =&gt; feasibility
    [description] =&gt; 
)
  --></li>
<li ><span class="name">generate and test</span> - See <span class="has_content"><a href="https://alandix.com/glossary/aibook/generate and test">generate and test</a></span><!--  Array
(
    [slug] =&gt; search!generate and test
    [parent] =&gt; search
    [name] =&gt; generate and test
    [canonical] =&gt; 
    [see] =&gt; generate and test
    [description] =&gt; 
)
  --></li>
<li ><span class="name"><a href="https://alandix.com/glossary/aibook/search!graph">graph</a></span><!--  Array
(
    [slug] =&gt; search!graph
    [parent] =&gt; search
    [name] =&gt; graph
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; 
)
  --></li>
<li ><span class="name"><a href="https://alandix.com/glossary/aibook/search!heuristic">heuristic</a></span><!--  Array
(
    [slug] =&gt; search!heuristic
    [parent] =&gt; search
    [name] =&gt; heuristic
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; 
)
  --></li>
<li ><span class="name"><a href="https://alandix.com/glossary/aibook/search!open list">open list</a></span><!--  Array
(
    [slug] =&gt; search!open list
    [parent] =&gt; search
    [name] =&gt; open list
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; 
)
  --></li>
<li ><span class="name"><a href="https://alandix.com/glossary/aibook/search!optimality">optimality</a></span><!--  Array
(
    [slug] =&gt; search!optimality
    [parent] =&gt; search
    [name] =&gt; optimality
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; 
)
  --></li>
<li ><span class="name"><a href="https://alandix.com/glossary/aibook/search!optimization">optimization</a></span><!--  Array
(
    [slug] =&gt; search!optimization
    [parent] =&gt; search
    [name] =&gt; optimization
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; 
)
  --></li>
<li ><span class="name"><a href="https://alandix.com/glossary/aibook/search!satisficing">satisficing</a></span><!--  Array
(
    [slug] =&gt; search!satisficing
    [parent] =&gt; search
    [name] =&gt; satisficing
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; 
)
  --></li>
<li ><span class="name"><a href="https://alandix.com/glossary/aibook/search!solution path">solution path</a></span><!--  Array
(
    [slug] =&gt; search!solution path
    [parent] =&gt; search
    [name] =&gt; solution path
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; 
)
  --></li>
<li ><span class="name"><a href="https://alandix.com/glossary/aibook/search!solution state">solution state</a></span><!--  Array
(
    [slug] =&gt; search!solution state
    [parent] =&gt; search
    [name] =&gt; solution state
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; 
)
  --></li></ul><!--  Array
(
    [slug] =&gt; search
    [parent] =&gt; 
    [name] =&gt; search
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; 
    [subentries] =&gt; Array
        (
            [search!adversarial] =&gt; Array
                (
                    [slug] =&gt; search!adversarial
                    [parent] =&gt; search
                    [name] =&gt; adversarial
                    [canonical] =&gt; 
                    [see] =&gt; 
                    [description] =&gt; 
                )

            [search!best first] =&gt; Array
                (
                    [slug] =&gt; search!best first
                    [parent] =&gt; search
                    [name] =&gt; best first
                    [canonical] =&gt; 
                    [see] =&gt; best first search
                    [description] =&gt; 
                )

            [search!breadth first] =&gt; Array
                (
                    [slug] =&gt; search!breadth first
                    [parent] =&gt; search
                    [name] =&gt; breadth first
                    [canonical] =&gt; 
                    [see] =&gt; breadth first search
                    [description] =&gt; 
                )

            [search!closed list] =&gt; Array
                (
                    [slug] =&gt; search!closed list
                    [parent] =&gt; search
                    [name] =&gt; closed list
                    [canonical] =&gt; 
                    [see] =&gt; closed list
                    [description] =&gt; 
                )

            [search!depth first] =&gt; Array
                (
                    [slug] =&gt; search!depth first
                    [parent] =&gt; search
                    [name] =&gt; depth first
                    [canonical] =&gt; 
                    [see] =&gt; depth first search
                    [description] =&gt; 
                )

            [search!deterministic] =&gt; Array
                (
                    [slug] =&gt; search!deterministic
                    [parent] =&gt; search
                    [name] =&gt; deterministic
                    [canonical] =&gt; 
                    [see] =&gt; deterministic search
                    [description] =&gt; 
                )

            [search!feasibility] =&gt; Array
                (
                    [slug] =&gt; search!feasibility
                    [parent] =&gt; search
                    [name] =&gt; feasibility
                    [canonical] =&gt; 
                    [see] =&gt; feasibility
                    [description] =&gt; 
                )

            [search!generate and test] =&gt; Array
                (
                    [slug] =&gt; search!generate and test
                    [parent] =&gt; search
                    [name] =&gt; generate and test
                    [canonical] =&gt; 
                    [see] =&gt; generate and test
                    [description] =&gt; 
                )

            [search!graph] =&gt; Array
                (
                    [slug] =&gt; search!graph
                    [parent] =&gt; search
                    [name] =&gt; graph
                    [canonical] =&gt; 
                    [see] =&gt; 
                    [description] =&gt; 
                )

            [search!heuristic] =&gt; Array
                (
                    [slug] =&gt; search!heuristic
                    [parent] =&gt; search
                    [name] =&gt; heuristic
                    [canonical] =&gt; 
                    [see] =&gt; 
                    [description] =&gt; 
                )

            [search!open list] =&gt; Array
                (
                    [slug] =&gt; search!open list
                    [parent] =&gt; search
                    [name] =&gt; open list
                    [canonical] =&gt; 
                    [see] =&gt; 
                    [description] =&gt; 
                )

            [search!optimality] =&gt; Array
                (
                    [slug] =&gt; search!optimality
                    [parent] =&gt; search
                    [name] =&gt; optimality
                    [canonical] =&gt; 
                    [see] =&gt; 
                    [description] =&gt; 
                )

            [search!optimization] =&gt; Array
                (
                    [slug] =&gt; search!optimization
                    [parent] =&gt; search
                    [name] =&gt; optimization
                    [canonical] =&gt; 
                    [see] =&gt; 
                    [description] =&gt; 
                )

            [search!satisficing] =&gt; Array
                (
                    [slug] =&gt; search!satisficing
                    [parent] =&gt; search
                    [name] =&gt; satisficing
                    [canonical] =&gt; 
                    [see] =&gt; 
                    [description] =&gt; 
                )

            [search!solution path] =&gt; Array
                (
                    [slug] =&gt; search!solution path
                    [parent] =&gt; search
                    [name] =&gt; solution path
                    [canonical] =&gt; 
                    [see] =&gt; 
                    [description] =&gt; 
                )

            [search!solution state] =&gt; Array
                (
                    [slug] =&gt; search!solution state
                    [parent] =&gt; search
                    [name] =&gt; solution state
                    [canonical] =&gt; 
                    [see] =&gt; 
                    [description] =&gt; 
                )

        )

)
  --></li>
  <li ><span class="name"><a href="https://alandix.com/glossary/aibook/search engines">search engines</a></span><!--  Array
(
    [slug] =&gt; search engines
    [parent] =&gt; 
    [name] =&gt; search engines
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; 
)
  --></li>
  <li ><span class="name"><a href="https://alandix.com/glossary/aibook/search horizon">search horizon</a></span><!--  Array
(
    [slug] =&gt; search horizon
    [parent] =&gt; 
    [name] =&gt; search horizon
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; 
)
  --></li>
  <li ><span class="name"><a href="https://alandix.com/glossary/aibook/search tree">search tree</a></span>- has subentries<ul class="subentries"><li ><span class="name"><a href="https://alandix.com/glossary/aibook/search tree!depth">depth</a></span><!--  Array
(
    [slug] =&gt; search tree!depth
    [parent] =&gt; search tree
    [name] =&gt; depth
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; 
)
  --></li></ul><!--  Array
(
    [slug] =&gt; search tree
    [parent] =&gt; 
    [name] =&gt; search tree
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; 
    [subentries] =&gt; Array
        (
            [search tree!depth] =&gt; Array
                (
                    [slug] =&gt; search tree!depth
                    [parent] =&gt; search tree
                    [name] =&gt; depth
                    [canonical] =&gt; 
                    [see] =&gt; 
                    [description] =&gt; 
                )

        )

)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/aibook/searle, john">Searle, John</a></span><!--  Array
(
    [slug] =&gt; searle, john
    [parent] =&gt; 
    [name] =&gt; Searle, John
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; John Searle is a philosopher famous within AI for his {{Chinese room argument}}, which holds that true machine intelligence is not possible because there is no {{intentlonality}} in a machine.,, words mean nothing to it.
)
  --></li>
  <li ><span class="name"><a href="https://alandix.com/glossary/aibook/seasonal adjustment">seasonal adjustment</a></span><!--  Array
(
    [slug] =&gt; seasonal adjustment
    [parent] =&gt; 
    [name] =&gt; seasonal adjustment
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; 
)
  --></li>
  <li ><span class="name"><a href="https://alandix.com/glossary/aibook/seasonal adjustments">seasonal adjustments</a></span><!--  Array
(
    [slug] =&gt; seasonal adjustments
    [parent] =&gt; 
    [name] =&gt; seasonal adjustments
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; 
)
  --></li>
  <li ><span class="name"><a href="https://alandix.com/glossary/aibook/segmentation">segmentation</a></span><!--  Array
(
    [slug] =&gt; segmentation
    [parent] =&gt; 
    [name] =&gt; segmentation
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; 
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/aibook/self learning">self learning</a></span><!--  Array
(
    [slug] =&gt; self learning
    [parent] =&gt; 
    [name] =&gt; self learning
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; Most {{machine learning}} is based on using large volumes of human-produced data.  This is even true of the way most {{unsupervised learning}} is used, as the data comes from previous human activity or data collection, such as medical records.  However, as models get ever larger, the available training data becomes a major bottleneck..   &lt;em&gt;Self-learning&lt;/em&gt; systems seek to go beyond the avaiable data in soem cases the AI system gathers its own data through experimentation, as in {{reinforcement learning}}, or where AI sysems are pitted against each other, as found in {{self play}} and {{adversarial learning}}.  These processes may be {{bootstrapped}} using initial human-created data (as in {{AlphaGo}} or work entirely from scratch using some form of specification of the problems (as in {{AlphaZero}}).
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/aibook/self play">self play</a></span><!--  Array
(
    [slug] =&gt; self play
    [parent] =&gt; 
    [name] =&gt; self play
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; Self play is used in {{gamp playing}} AI in order to grow the available data of games.  Game playing systems can be trained using records of past human games, for example, extensive records of chess competitions,  However, this apprach is limited by the available data.  Systems such as {{AlphaGo}} and its successors grew their available data by pitting AI aganst AI, this being able to create vast numbers of entirely machine-generated games.  This data can then be used to train better game playing AI, and these used to generate yet more games.
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/aibook/self-organising map">self-organising map</a></span><!--  Array
(
    [slug] =&gt; self-organising map
    [parent] =&gt; 
    [name] =&gt; self-organising map
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; A self-organising map (SOM) is a form of {{unsupervised learning}} where nodes on a 2D grid (or more geometric lattice structure) act as archetypes or key values for a form of clustering, where like items are allocated to grid points that have some form of relationship.    The prime example of a SOM are {{Kohonen networks}}.  Like other forms of {{clustering algorithm}}, SOM can be used for data interpretation where a human expert labels areas on the map, or as a form of {[dimension reduction}} either using the nodes as discrete {{categorical}} clusters or using the 2D (or higer diennsional) embedding of the grid or lattice as {{continuous}} dimensions.
)
  --></li>
  <li ><span class="name">self-organizing network</span> - See <span class="has_content"><a href="https://alandix.com/glossary/aibook/self-organising map">self-organising map</a></span><!--  Array
(
    [slug] =&gt; self-organizing network
    [parent] =&gt; 
    [name] =&gt; self-organizing network
    [canonical] =&gt; 
    [see] =&gt; self-organising map
    [description] =&gt; 
)
  --></li>
  <li ><span class="name"><a href="https://alandix.com/glossary/aibook/semantic analysis">semantic analysis</a></span><!--  Array
(
    [slug] =&gt; semantic analysis
    [parent] =&gt; 
    [name] =&gt; semantic analysis
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; 
)
  --></li>
  <li ><span class="name"><a href="https://alandix.com/glossary/aibook/semantic desktop">semantic desktop</a></span><!--  Array
(
    [slug] =&gt; semantic desktop
    [parent] =&gt; 
    [name] =&gt; semantic desktop
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; 
)
  --></li>
  <li ><span class="name"><a href="https://alandix.com/glossary/aibook/semantic grammar">semantic grammar</a></span><!--  Array
(
    [slug] =&gt; semantic grammar
    [parent] =&gt; 
    [name] =&gt; semantic grammar
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; 
)
  --></li>
  <li ><span class="name"><a href="https://alandix.com/glossary/aibook/semantic integrity">semantic integrity</a></span><!--  Array
(
    [slug] =&gt; semantic integrity
    [parent] =&gt; 
    [name] =&gt; semantic integrity
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; 
)
  --></li>
  <li ><span class="name"><a href="https://alandix.com/glossary/aibook/semantic network">semantic network</a></span><!--  Array
(
    [slug] =&gt; semantic network
    [parent] =&gt; 
    [name] =&gt; semantic network
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; 
)
  --></li>
  <li ><span class="name"><a href="https://alandix.com/glossary/aibook/semantic web">semantic web</a></span><!--  Array
(
    [slug] =&gt; semantic web
    [parent] =&gt; 
    [name] =&gt; semantic web
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; 
)
  --></li>
  <li ><span class="name"><a href="https://alandix.com/glossary/aibook/semi-autonomous car">semi-autonomous car</a></span><!--  Array
(
    [slug] =&gt; semi-autonomous car
    [parent] =&gt; 
    [name] =&gt; semi-autonomous car
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; 
)
  --></li>
  <li ><span class="name"><a href="https://alandix.com/glossary/aibook/semi-supervised learning">semi-supervised learning</a></span><!--  Array
(
    [slug] =&gt; semi-supervised learning
    [parent] =&gt; 
    [name] =&gt; semi-supervised learning
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; 
)
  --></li>
  <li ><span class="name"><a href="https://alandix.com/glossary/aibook/sensation">sensation</a></span><!--  Array
(
    [slug] =&gt; sensation
    [parent] =&gt; 
    [name] =&gt; sensation
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; 
)
  --></li>
  <li ><span class="name"><a href="https://alandix.com/glossary/aibook/senses">senses</a></span><!--  Array
(
    [slug] =&gt; senses
    [parent] =&gt; 
    [name] =&gt; senses
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; 
)
  --></li>
  <li ><span class="name"><a href="https://alandix.com/glossary/aibook/sensitivity">sensitivity</a></span><!--  Array
(
    [slug] =&gt; sensitivity
    [parent] =&gt; 
    [name] =&gt; sensitivity
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; 
)
  --></li>
  <li ><span class="name"><a href="https://alandix.com/glossary/aibook/sensor fusion">sensor fusion</a></span><!--  Array
(
    [slug] =&gt; sensor fusion
    [parent] =&gt; 
    [name] =&gt; sensor fusion
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; 
)
  --></li>
  <li ><span class="name"><a href="https://alandix.com/glossary/aibook/sentence parsing">sentence parsing</a></span>- has subentries<ul class="subentries"><li ><span class="name"><a href="https://alandix.com/glossary/aibook/sentence parsing!bottom up">bottom up</a></span><!--  Array
(
    [slug] =&gt; sentence parsing!bottom up
    [parent] =&gt; sentence parsing
    [name] =&gt; bottom up
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; 
)
  --></li>
<li ><span class="name"><a href="https://alandix.com/glossary/aibook/sentence parsing!methods">methods</a></span><!--  Array
(
    [slug] =&gt; sentence parsing!methods
    [parent] =&gt; sentence parsing
    [name] =&gt; methods
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; 
)
  --></li>
<li ><span class="name"><a href="https://alandix.com/glossary/aibook/sentence parsing!systems">systems</a></span><!--  Array
(
    [slug] =&gt; sentence parsing!systems
    [parent] =&gt; sentence parsing
    [name] =&gt; systems
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; 
)
  --></li>
<li ><span class="name"><a href="https://alandix.com/glossary/aibook/sentence parsing!top down">top down</a></span><!--  Array
(
    [slug] =&gt; sentence parsing!top down
    [parent] =&gt; sentence parsing
    [name] =&gt; top down
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; 
)
  --></li></ul><!--  Array
(
    [slug] =&gt; sentence parsing
    [parent] =&gt; 
    [name] =&gt; sentence parsing
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; 
    [subentries] =&gt; Array
        (
            [sentence parsing!bottom up] =&gt; Array
                (
                    [slug] =&gt; sentence parsing!bottom up
                    [parent] =&gt; sentence parsing
                    [name] =&gt; bottom up
                    [canonical] =&gt; 
                    [see] =&gt; 
                    [description] =&gt; 
                )

            [sentence parsing!methods] =&gt; Array
                (
                    [slug] =&gt; sentence parsing!methods
                    [parent] =&gt; sentence parsing
                    [name] =&gt; methods
                    [canonical] =&gt; 
                    [see] =&gt; 
                    [description] =&gt; 
                )

            [sentence parsing!systems] =&gt; Array
                (
                    [slug] =&gt; sentence parsing!systems
                    [parent] =&gt; sentence parsing
                    [name] =&gt; systems
                    [canonical] =&gt; 
                    [see] =&gt; 
                    [description] =&gt; 
                )

            [sentence parsing!top down] =&gt; Array
                (
                    [slug] =&gt; sentence parsing!top down
                    [parent] =&gt; sentence parsing
                    [name] =&gt; top down
                    [canonical] =&gt; 
                    [see] =&gt; 
                    [description] =&gt; 
                )

        )

)
  --></li>
  <li ><span class="name"><a href="https://alandix.com/glossary/aibook/sentence parsing methods">sentence parsing methods</a></span><!--  Array
(
    [slug] =&gt; sentence parsing methods
    [parent] =&gt; 
    [name] =&gt; sentence parsing methods
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; 
)
  --></li>
  <li ><span class="name"><a href="https://alandix.com/glossary/aibook/sentence-level processing">sentence-level processing</a></span><!--  Array
(
    [slug] =&gt; sentence-level processing
    [parent] =&gt; 
    [name] =&gt; sentence-level processing
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; 
)
  --></li>
  <li ><span class="name"><a href="https://alandix.com/glossary/aibook/sequence order">sequence order</a></span><!--  Array
(
    [slug] =&gt; sequence order
    [parent] =&gt; 
    [name] =&gt; sequence order
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; 
)
  --></li>
  <li ><span class="name"><a href="https://alandix.com/glossary/aibook/set theory">set theory</a></span><!--  Array
(
    [slug] =&gt; set theory
    [parent] =&gt; 
    [name] =&gt; set theory
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; 
)
  --></li>
  <li ><span class="name"><a href="https://alandix.com/glossary/aibook/sexual selection">sexual selection</a></span><!--  Array
(
    [slug] =&gt; sexual selection
    [parent] =&gt; 
    [name] =&gt; sexual selection
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; 
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/aibook/sharding">sharding</a></span><!--  Array
(
    [slug] =&gt; sharding
    [parent] =&gt; 
    [name] =&gt; sharding
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; Sharding is used in {{big data}} applications to divide the data into portions (called shards) so that they can be stored and processed semi-independently.  If we think of simple tabel data, the shards may be split along the rows, so that all the data for a single item (row) is always stored together.  Alternatively particular columns (item fields/features) may be stored sepaartely, linked by some form of {{key}} or {{identifier}}.  The nature of the sharding affects the efficiency of algorithms, so there is often a desire to put data that needs to be processed together in the same shard.  This is particularly difficult for {{graph data}}, such as {{social network data}}, as the links between items inevitably cut across shards.
)
  --></li>
  <li ><span class="name"><a href="https://alandix.com/glossary/aibook/sheep dogs">sheep dogs</a></span><!--  Array
(
    [slug] =&gt; sheep dogs
    [parent] =&gt; 
    [name] =&gt; sheep dogs
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; 
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/aibook/shelley, mary">Shelley, Mary</a></span><!--  Array
(
    [slug] =&gt; shelley, mary
    [parent] =&gt; 
    [name] =&gt; Shelley, Mary
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; Mary Shelley is the author of the story of {{Frankenstein}}'s monster.
)
  --></li>
  <li ><span class="name"><a href="https://alandix.com/glossary/aibook/shrdlu">SHRDLU</a></span><!--  Array
(
    [slug] =&gt; shrdlu
    [parent] =&gt; 
    [name] =&gt; SHRDLU
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; 
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/aibook/sigmoid activation function">sigmoid activation function</a></span><!--  Array
(
    [slug] =&gt; sigmoid activation function
    [parent] =&gt; 
    [name] =&gt; sigmoid activation function
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; A sigmoid activation function is used in many {{neural networks}}.  the properties of a {{sigmoid function}} mean that it shares some of the {{non-lonearity}} of a {{threshold function}}, but is {{differentiable}} enabling learning through the {{backpropogation algorithm}}.
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/aibook/sigmoid function">sigmoid function</a></span><!--  Array
(
    [slug] =&gt; sigmoid function
    [parent] =&gt; 
    [name] =&gt; sigmoid function
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; A sigmoid function is a generic term refering to S-like functions that have a step-like nature, but are smooth without the {{discontinuity}} of the {{threshold function}}.  This maes them more emnable to use in {{machine learning}} as, in general, {{continuity}} aids learning.
)
  --></li>
  <li ><span class="name"><a href="https://alandix.com/glossary/aibook/signal processing">signal processing</a></span><!--  Array
(
    [slug] =&gt; signal processing
    [parent] =&gt; 
    [name] =&gt; signal processing
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; 
)
  --></li>
  <li ><span class="name"><a href="https://alandix.com/glossary/aibook/similarity">similarity</a></span><!--  Array
(
    [slug] =&gt; similarity
    [parent] =&gt; 
    [name] =&gt; similarity
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; 
)
  --></li>
  <li ><span class="name"><a href="https://alandix.com/glossary/aibook/similarity matrix">similarity matrix</a></span><!--  Array
(
    [slug] =&gt; similarity matrix
    [parent] =&gt; 
    [name] =&gt; similarity matrix
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; 
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/aibook/simon, herbert">Simon, Herbert</a></span><!--  Array
(
    [slug] =&gt; simon, herbert
    [parent] =&gt; 
    [name] =&gt; Simon, Herbert
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; Herbert A. Simon was a political economist and AI pioneer who, amnstt other things, coned the terms {{satisficing}} and {{bounded rationality}}.  He also, with {{Alan Newell}}, proposed the {{physical symbol system hypothesis}}.
)
  --></li>
  <li ><span class="name"><a href="https://alandix.com/glossary/aibook/simulated annealing">simulated annealing</a></span><!--  Array
(
    [slug] =&gt; simulated annealing
    [parent] =&gt; 
    [name] =&gt; simulated annealing
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; 
)
  --></li>
  <li ><span class="name"><a href="https://alandix.com/glossary/aibook/simulated data">simulated data</a></span><!--  Array
(
    [slug] =&gt; simulated data
    [parent] =&gt; 
    [name] =&gt; simulated data
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; 
)
  --></li>
  <li ><span class="name"><a href="https://alandix.com/glossary/aibook/single-shot learning">single-shot learning</a></span><!--  Array
(
    [slug] =&gt; single-shot learning
    [parent] =&gt; 
    [name] =&gt; single-shot learning
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; 
)
  --></li>
  <li ><span class="name"><a href="https://alandix.com/glossary/aibook/singularity">singularity</a></span><!--  Array
(
    [slug] =&gt; singularity
    [parent] =&gt; 
    [name] =&gt; singularity
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; 
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/aibook/skynet project">Skynet</a></span><!--  Array
(
    [slug] =&gt; skynet project
    [parent] =&gt; 
    [name] =&gt; Skynet
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; Skynet is the name for a system of CCTV cameras, {{facial recogntion}} technlogy and {{biometric}} databases deployed in mainland China.  It is used for criminal investigations, but also for political purposes.  Skynet is of course also the name of the comouter system that destroys most of humankind in the Terminator films.
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/aibook/smart dust">smart dust</a></span><!--  Array
(
    [slug] =&gt; smart dust
    [parent] =&gt; 
    [name] =&gt; smart dust
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; Smart dust refers to tiny sensing and processing units designed to be scattered in large numbers, typically for sensing purposes. Each is individually poor fidelity but collectively they can deliver important information. Smart dust usually has some form of network communication to enable collective processing, but this needs to be low bandwidt and range because of the mited power availability of the devices.  The main uses are miitary and environmental (although the smart device units themselves are polluting if not recovered).
)
  --></li>
  <li ><span class="name"><a href="https://alandix.com/glossary/aibook/smoothing">smoothing</a></span><!--  Array
(
    [slug] =&gt; smoothing
    [parent] =&gt; 
    [name] =&gt; smoothing
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; 
)
  --></li>
  <li ><span class="name"><a href="https://alandix.com/glossary/aibook/smoothing function">smoothing function</a></span><!--  Array
(
    [slug] =&gt; smoothing function
    [parent] =&gt; 
    [name] =&gt; smoothing function
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; 
)
  --></li>
  <li ><span class="name"><a href="https://alandix.com/glossary/aibook/snipit">Snip!t</a></span><!--  Array
(
    [slug] =&gt; snipit
    [parent] =&gt; 
    [name] =&gt; Snip!t
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; 
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/aibook/soar">SOAR</a></span><!--  Array
(
    [slug] =&gt; soar
    [parent] =&gt; 
    [name] =&gt; SOAR
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; SOAR is a {{cognitive architecture}} crated by {{John Laird}}, {{Alan Newell}} and {{Paul Rosenbloom}}.
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/aibook/sobel's operator">Sobel's operator</a></span><!--  Array
(
    [slug] =&gt; sobel's operator
    [parent] =&gt; 
    [name] =&gt; Sobel's operator
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; Sobel'soperator is used to assess whether there is a line at a particular location and to estmiate its direction.  It can be seen as the sum of two factors, one measuring the horizontal rate of change and the other the virtical rate of change. &lt;br&gt;&lt;br&gt; &amp;nbsp; &amp;nbsp; &amp;nbsp; H &amp;nbsp; = &amp;nbsp; (c+2f+i) - (a+2d+g)  &lt;br&gt;
&amp;nbsp; &amp;nbsp; &amp;nbsp; V &amp;nbsp; = &amp;nbsp; (g+2h+i) - (a+2b+c)  &lt;br&gt;
&amp;nbsp; &amp;nbsp; &amp;nbsp; G  &amp;nbsp; = &amp;nbsp; | H |  +  | V | &lt;br&gt; where ,a,b,c,d,e,f,g,h,i are the pixels in a 3x3 square (see diagram)
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/aibook/social machine">social machine</a></span><!--  Array
(
    [slug] =&gt; social machine
    [parent] =&gt; 
    [name] =&gt; social machine
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; Those working in {{web science}} ofter refer to the web as a 'social machine' as the human users and authors of the web through their social interactions captured in the web (e.g. through social media) and those mediated by it (e.g. publishing pages for others to view) can be regarded as collectively processing.
)
  --></li>
  <li ><span class="name"><a href="https://alandix.com/glossary/aibook/social media">social media</a></span><!--  Array
(
    [slug] =&gt; social media
    [parent] =&gt; 
    [name] =&gt; social media
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; 
)
  --></li>
  <li ><span class="name"><a href="https://alandix.com/glossary/aibook/software agents">software agents</a></span><!--  Array
(
    [slug] =&gt; software agents
    [parent] =&gt; 
    [name] =&gt; software agents
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; 
)
  --></li>
  <li ><span class="name"><a href="https://alandix.com/glossary/aibook/software ring-fence">software ring-fence</a></span><!--  Array
(
    [slug] =&gt; software ring-fence
    [parent] =&gt; 
    [name] =&gt; software ring-fence
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; 
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/aibook/sparql">SPARQL</a></span><!--  Array
(
    [slug] =&gt; sparql
    [parent] =&gt; 
    [name] =&gt; SPARQL
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; SPARQL is a query language for {{RDF}} data.  It is a form of {{graph query}} for triple-based representations.
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/aibook/sparse matrix">sparse matrix</a></span><!--  Array
(
    [slug] =&gt; sparse matrix
    [parent] =&gt; 
    [name] =&gt; sparse matrix
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; A {{matrix}} is called sparse if most of its entries are zero (or empty).  Sparse matrices can be store efficiently by only recording the entroes with actual values.  However, this can make algorithms to deal with sparse matrices more complicated.
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/aibook/spectrogram">spectrogram</a></span><!--  Array
(
    [slug] =&gt; spectrogram
    [parent] =&gt; 
    [name] =&gt; spectrogram
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; A spectrogram shows a sound, such as speach, based on its frequency components. 
)
  --></li>
  <li ><span class="name"><a href="https://alandix.com/glossary/aibook/speech acts">speech acts</a></span><!--  Array
(
    [slug] =&gt; speech acts
    [parent] =&gt; 
    [name] =&gt; speech acts
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; 
)
  --></li>
  <li ><span class="name"><a href="https://alandix.com/glossary/aibook/speech recognition">speech recognition</a></span><!--  Array
(
    [slug] =&gt; speech recognition
    [parent] =&gt; 
    [name] =&gt; speech recognition
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; 
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/aibook/spikes">spikes</a></span><!--  Array
(
    [slug] =&gt; spikes
    [parent] =&gt; 
    [name] =&gt; spikes
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; When {{biological neurones}} pass potential to one another, they do not do so in lock step and using continuous vvalues, as is usually the case in {[artificial neutal neteotks}}, but as short spikes.  These are more likely to occur when a cell is highly activated, leading to more frequent spikes, and the continuous value approxmates this, but the spiking cam lead to differet behaviours.  This more detail bilogval model is te bases if {{spiking neural networks}}.
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/aibook/spiking neural networks">spiking neural networks</a></span><!--  Array
(
    [slug] =&gt; spiking neural networks
    [parent] =&gt; 
    [name] =&gt; spiking neural networks
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; A spiking neural network emulates the way that real {{neurones}} in the brain communicate through discrete {{spikes}} rather than weighted links as is most common in {{artificial neural networks}}.  In some ways the two behave similarly and a weighted impulse is rougly equivalent to a fast firing spike link.  However, there are differences, for example the potential for {{disambiguation}}.
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/aibook/spin glass models">spin glass models</a></span><!--  Array
(
    [slug] =&gt; spin glass models
    [parent] =&gt; 
    [name] =&gt; spin glass models
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; Spin glass models are used to physics to explain the way that magnetic materials that initially have spins (generating their magnetis moment) in random directiosn, settle down to have spins aligned over patches or the entire material.  They are used as an analogy for {{subsymbolic}} {{machine learning}} including {{Bolzmann machines}} and {{Hopfield networks}}.
)
  --></li>
  <li ><span class="name"><a href="https://alandix.com/glossary/aibook/sporadic">sporadic</a></span><!--  Array
(
    [slug] =&gt; sporadic
    [parent] =&gt; 
    [name] =&gt; sporadic
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; 
)
  --></li>
  <li ><span class="name"><a href="https://alandix.com/glossary/aibook/spreading activation">spreading activation</a></span><!--  Array
(
    [slug] =&gt; spreading activation
    [parent] =&gt; 
    [name] =&gt; spreading activation
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; 
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/aibook/square matrices">square matrices</a></span><!--  Array
(
    [slug] =&gt; square matrices
    [parent] =&gt; 
    [name] =&gt; square matrices
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; A square matrix is a {{matrix}} with the same number of rows and columns.
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/aibook/standard logic">standard logic</a></span><!--  Array
(
    [slug] =&gt; standard logic
    [parent] =&gt; 
    [name] =&gt; standard logic
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; Standard logic uses simple facts and rules abiut a single situation.  For example, the classic &quot;Socrates is a man; all men are mortal, therefore Socrates is mortal&quot;.   This example includes named  things (Socrates), {{predicates}} (properties of things, such as &quot;is a man&quot;), of {{propositions}} (facts such as &quot;Socrates is a man&quot;),  {{quantifiers}} (ways talking about lots of things at once, such as &quot;all&quot;) and {{syllogisms}} (reasoning).   These are often written in formal terms such as: &lt;br&gt; &amp;nbsp; &amp;nbsp; &amp;nbsp; is_a_man(Socrates)   &lt;br&gt; &amp;nbsp; &amp;nbsp; &amp;nbsp; &amp;forall; p :  is_a_man(p) &amp;Implies;   mortal(p)  &lt;br&gt; &amp;nbsp; &amp;nbsp; &amp;nbsp; mortal(Socrates)  &amp;nbsp; &amp;nbsp; -- from the rule  ( &amp;forall; x :  P(x) &amp; Implies;   Q(x)  ) and P(y) infer  Q(y)  &lt;br&gt; Note how there are two levels of implication here the implies in the propositions (&amp;Implies; ) and the rule implication.  The former that can be regarded as a {{syntactic sugar}}  for the equivaent ands and ors: &lt;br&gt; &amp;nbsp; &amp;nbsp; &amp;nbsp; X &amp;Implies; Y. &amp;equiv;  &amp;not; X &amp;or; Y  &lt;br&gt; &lt;br&gt; Standard logics  can be contrasted with {{non-standard logic}} such as {{non-monotonic reasoning}}, {{modal logic}} or {{fuzzy logic}}
)
  --></li>
  <li ><span class="name"><a href="https://alandix.com/glossary/aibook/state space search">state space search</a></span><!--  Array
(
    [slug] =&gt; state space search
    [parent] =&gt; 
    [name] =&gt; state space search
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; 
)
  --></li>
  <li ><span class="name"><a href="https://alandix.com/glossary/aibook/state space traversal">state space traversal</a></span><!--  Array
(
    [slug] =&gt; state space traversal
    [parent] =&gt; 
    [name] =&gt; state space traversal
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; 
)
  --></li>
  <li ><span class="name"><a href="https://alandix.com/glossary/aibook/statistical techniques">statistical techniques</a></span><!--  Array
(
    [slug] =&gt; statistical techniques
    [parent] =&gt; 
    [name] =&gt; statistical techniques
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; 
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/aibook/status change event">status change event</a></span><!--  Array
(
    [slug] =&gt; status change event
    [parent] =&gt; 
    [name] =&gt; status change event
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; A status change event is an {{event}} that is registered when some sensor or digital value changes.  For example, a change in the temeprature in a central heating systems, a finger movement in a {{user interface}}, or a share price .on a stock exchange.  The event may be used to {{trigger}} an action in a software {{agent}} or rase a {{notification}} for a user.  Discrete changes in status, such as an in/out indicator on an office door may have an idetifiable event that causes the change, and thus this can be monitored.  However sometime other means such as periodic {{polling}} are necessary.  For continuous values often very small changes are ignored and the status change event only raised wen the change exceeds some {{threshold}}.
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/aibook/stemming">stemming</a></span><!--  Array
(
    [slug] =&gt; stemming
    [parent] =&gt; 
    [name] =&gt; stemming
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; Stemming is a process of reducing a word to its stem as a normalised form for the purposes of {{natural language processing}}, {{indexing}} or {{document retreval}}.  For example, plurals are reduced to their singular form by deleting the trailing 's' and gerunds of verbs have theor 'ing' removed.  Stemming is usually perfomed at a {{lexical}} level without any {{semantic knowledge}}, which can lead to a small proportion of inappropriate stems, for example,  'starling' might be stemmed to 'starl'.
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/aibook/stereo vision">stereo vision</a></span><!--  Array
(
    [slug] =&gt; stereo vision
    [parent] =&gt; 
    [name] =&gt; stereo vision
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; Stereo vision uses two (or more) cameras in order, inter alia, to be able to detect the distance of objects using {{triangulation}}.
)
  --></li>
  <li ><span class="name"><a href="https://alandix.com/glossary/aibook/stimulus-response learning">stimulus-response learning</a></span><!--  Array
(
    [slug] =&gt; stimulus-response learning
    [parent] =&gt; 
    [name] =&gt; stimulus-response learning
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; 
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/aibook/stop words">stop words</a></span><!--  Array
(
    [slug] =&gt; stop words
    [parent] =&gt; 
    [name] =&gt; stop words
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; In {{natural language processing}} stop words are words that are removed or ignored before further processing.  They are usually the most common words in language such as 'the', 'a', or 'is'.
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/aibook/strong ai">strong AI</a></span><!--  Array
(
    [slug] =&gt; strong ai
    [parent] =&gt; 
    [name] =&gt; strong AI
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; Strong AI is the belief that computers can (and will) be able to achieve full {{human-like intelligence}}, in part because human intelligence itself, so it would be claimed, is effectively built on a biological computer, the brain.  Strong AI stands in contrast to {{weak AI}} which only seeks computer systems that perform in an intelligent manner even if they do so in very different ways to humans.
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/aibook/strong learner">strong learner</a></span><!--  Array
(
    [slug] =&gt; strong learner
    [parent] =&gt; 
    [name] =&gt; strong learner
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; In {{machine learner}} a strong learner is a model that on ist own is suffciently effective to be used for all data.  It is in contrast to a {{weak learner}} that in some way would need to be combined with ither weak learners to be useful.
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/aibook/stuttering in search trees">stuttering in search trees</a></span><!--  Array
(
    [slug] =&gt; stuttering in search trees
    [parent] =&gt; 
    [name] =&gt; stuttering in search trees
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; Stuttering in search trees arises when moves in the {{search tree}} can go back and forth between the same states.  For example, in the {{Towers of Hanoi}} one could repeatedly move a ring form tower 1 to 2 on the forts movem then form 2 to one on the next move and so on.  A {{closed list}} can be used to record visited states and hence avoid stuttering.
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/aibook/stuxnet">Stuxnet</a></span><!--  Array
(
    [slug] =&gt; stuxnet
    [parent] =&gt; 
    [name] =&gt; Stuxnet
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; Stuxnet was the worlds first known true {{cyber-weapon}}.   Stuxnet was a {{worm}} transported on USB pen-drive sticks and designed to target centrifuges at Iranian nuclear processing facilities.  Due to its complexity, it was widely believed to have been a joint effort of security services in Israel and USA, who wished to prevent Iran obtaining nuclear material for weapons, but neither ever confirmed this.  It targetted parricular micro-processors that were used in the centrifuges and instructed them to drive the centrifuge motors in patterns that physically damaged them.  However, the same micro-processor was used in other kinds of equipment and there were unconformed reports that Stuxnet had been found in Ukrainian and Russian nuclear power stations.
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/aibook/sub-symbolic systems">sub-symbolic systems</a></span><!--  Array
(
    [slug] =&gt; sub-symbolic systems
    [parent] =&gt; 
    [name] =&gt; sub-symbolic systems
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; Subsymbolic systems are those which use low-level representations such as the weights in a {{neural newtworks}} or the artificial insects in {{swarm computing}}.  One could argue that there is still digital symbol manipulation when they execute, but not at a recognisable level.  In contrast {{symbolic systems}} use symbols such as names, places, and concepts whch are manipulated by explicit rules.
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/aibook/subject-predicate-object triples">subject--predicate--object triples</a></span><!--  Array
(
    [slug] =&gt; subject-predicate-object triples
    [parent] =&gt; 
    [name] =&gt; subject--predicate--object triples
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; Subject-predicate-object triples (or simply {{triples}}) are the main form of {{knowledge representation}} in {{RDF}}, the language of the {{semantic web}}.   Becasue of this {{RDF databases}} are often referd to has {{triple stores}}.  Although the {{triple}} seems verey simple, it can dscribe anything that more complex representations such as {{tables}}, {{XML}} or {{JSON data structures}}   Some {{RDF databases}} also stupport {{named graphs}}, where {{triples}} are tagged with a graph URI, and hence are called {{quad stores}}.&lt;br&gt;&lt;br&gt;An example: &lt;br&gt; &lt;table align=&quot;center&quot;&gt;&lt;tr&gt;&lt;th align=&quot;center&quot;&gt;subject&lt;/th&gt;&lt;td&gt;&lt;/td&gt;&lt;th align=&quot;center&quot;&gt;predicate&lt;/th&gt;&lt;td&gt;&lt;/td&gt;&lt;th align=&quot;center&quot;&gt;object&lt;/th&gt;&lt;/tr&gt;&lt;tr&gt;&lt;td align=&quot;center&quot;&gt;Scooby Doo&lt;/td&gt;&lt;td align=&quot;center&quot;&gt;&ndash;-&lt;/td&gt;&lt;td align=&quot;center&quot;&gt;instance of&lt;/td&gt;&lt;td align=&quot;center&quot;&gt;&ndash;-&lt;/td&gt;&lt;td align=&quot;center&quot;&gt;Great Dane&lt;/td&gt;&lt;/tr&gt;&lt;tr&gt;&lt;td align=&quot;center&quot;&gt;Dog&lt;/td&gt;&lt;td align=&quot;center&quot;&gt;&ndash;-&lt;/td&gt;&lt;td align=&quot;center&quot;&gt;number of legs&lt;/td&gt;&lt;td align=&quot;center&quot;&gt;&ndash;-&lt;/td&gt;&lt;td align=&quot;center&quot;&gt;4&lt;/td&gt;&lt;/tr&gt;&lt;/table&gt;&lt;br&gt;In {{RDF}} the subject and predicate must both be {{URI}}s, like this: &lt;br&gt; &lt;table align=&quot;center&quot;&gt;&lt;tr&gt;&lt;td align=&quot;center&quot;&gt;&amp;lt;https://alandix.com/&amp;gt;&lt;/td&gt;&lt;td align=&quot;center&quot;&gt;&amp;lt;dc:creator&amp;gt;&lt;/td&gt;           &lt;td align=&quot;center&quot;&gt;&quot;Alan Dix&quot;&lt;/td&gt;&lt;/tr&gt;
&lt;tr&gt;&lt;td align=&quot;center&quot;&gt;&amp;lt;aibk:Scooby_Doo&amp;gt;&lt;/td&gt;       &lt;td align=&quot;center&quot;&gt;&amp;lt;rdf:type&amp;gt;&lt;/td&gt;             &lt;td align=&quot;center&quot;&gt;&amp;lt;aibk:Great_Dane&amp;gt;&lt;/td&gt;&lt;/tr&gt;
&lt;tr&gt;&lt;td align=&quot;center&quot;&gt;&amp;lt;aibk:Dog&amp;gt;&lt;/td&gt;              &lt;td align=&quot;center&quot;&gt;&amp;lt;aibk:number_of_legs&amp;gt;&lt;/td&gt;  &lt;td align=&quot;center&quot;&gt;&quot;4&quot;^^xsd:integer&lt;/td&gt;&lt;/tr&gt;&lt;/table&gt;


&ndash; &ndash; 4&lt;/table&gt;
)
  --></li>
  <li ><span class="name"><a href="https://alandix.com/glossary/aibook/successive refinement">successive refinement</a></span><!--  Array
(
    [slug] =&gt; successive refinement
    [parent] =&gt; 
    [name] =&gt; successive refinement
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; 
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/aibook/sufficient reason">sufficient reason</a></span><!--  Array
(
    [slug] =&gt; sufficient reason
    [parent] =&gt; 
    [name] =&gt; sufficient reason
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; In human communication we only feel the need to give sufficient reason for our xdecisions or actions, not unpack the firing of {{neorones}} in our brains.  This principle can be applied in {{explainable AI}}, when it is tempting to try to explain the minutae of, say, every node and oink in a {{deep neural network}}.   Instead, we should seek explanations at an appropriate level for the task at hand.
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/aibook/supervised learning">supervised learning</a></span><!--  Array
(
    [slug] =&gt; supervised learning
    [parent] =&gt; 
    [name] =&gt; supervised learning
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; In supervised learning the {{training set}} is {{labelled}}, that is it has some form of markers to indicate a desired grouping or value.  The supervised algorithm typically attempts to emulate or approximate the labelling of the training set.
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/aibook/support vector machine">support vector machine</a></span><!--  Array
(
    [slug] =&gt; support vector machine
    [parent] =&gt; 
    [name] =&gt; support vector machine
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; A support vector machine (SVM) is a {{multi-layer neural network}} where the first layer uses a large number of specialised {{non-linear functions}} in order to allow simpler leaerning in later levels.   Often the non-linear functions form a family, called a {{kernal}}, for example they maybe foms of {{wavelet}}.  Alongside other techniques such as {{reservoir ocmputing}} it is a way of creating {{non-linear diversity}}.  
)
  --></li>
  <li ><span class="name"><a href="https://alandix.com/glossary/aibook/surrogate expert">surrogate expert</a></span><!--  Array
(
    [slug] =&gt; surrogate expert
    [parent] =&gt; 
    [name] =&gt; surrogate expert
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; 
)
  --></li>
  <li ><span class="name">svm</span> - See <span class="has_content"><a href="https://alandix.com/glossary/aibook/support vector machine">support vector machine</a></span><!--  Array
(
    [slug] =&gt; svm
    [parent] =&gt; 
    [name] =&gt; svm
    [canonical] =&gt; 
    [see] =&gt; support vector machine
    [description] =&gt; 
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/aibook/swarm computing">swarm computing</a></span><!--  Array
(
    [slug] =&gt; swarm computing
    [parent] =&gt; 
    [name] =&gt; swarm computing
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; Swarm computing is based on analogies with social insects such as ants and bees.  These insects behave in intelligent ways as a collective even though each individual insect is very simple.  In swarm computing many simple individual agents are prodused which explore a spece, for example to find an optmal solution.  They operate semi-independently with a level of simple communication that nudges their behaviour.  For examaple, if an agent in the swarm has found a good solution it may retrace the intermediate states it took to get there labelling them as good, but with decaying strengh, thus mirroring phermone trails.  Swarm computing is similar to {[ensemble method}}, but the latter are more about forming a diverse group of simple {[decsions making}} agents that together are used repeatedly to solve similar problems, wheres in swarm computing the agents are near identical and it usualy only the final result that matters.
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/aibook/syllogism">syllogism</a></span><!--  Array
(
    [slug] =&gt; syllogism
    [parent] =&gt; 
    [name] =&gt; syllogism
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; A syllogism is a formal rule to combine logical premises. The classic is {{Aristotle}}'s: &lt;br&gt; &amp;nbsp; FROM &quot;all men are mortal &lt;br&gt; &amp;nbsp; AND &quot;Socrates is a man&quot; &lt;br&gt; &amp;nbsp; CONCLUDE: &quot;Socrates is mortal&quot;. &lt;br&gt;. This is using a general syllogism of the form: &lt;br&gt; &amp;nbsp; &amp;nbsp; FROM &quot;all members of CLASS(x) HAS_PROPERTY(y) &lt;br&gt; &amp;nbsp; AND &quot;ENTITY(z) is of CLASS(x)&quot; &lt;br&gt; &amp;nbsp; CONCLUDE: &quot;ENTITY(z)  HAS_PROPERTY(y)&quot;
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/aibook/symbolic regression">symbolic regression</a></span><!--  Array
(
    [slug] =&gt; symbolic regression
    [parent] =&gt; 
    [name] =&gt; symbolic regression
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; Symbolic regression is a technique in machine learning that attempts to create succinct equations to describe data.  It can be used in physics or engineering to suggest potential physical laws from empirical data.  Symbolic regression is named to suggest the connection with {{linear regression}}.  Whilst {{linear regression}} can only create linear relationships such as y = 3x+5, symbolic regression can look for complex formulae such as F = q(E&lt;sub&gt;f&lt;/sub&gt;  + B v sin &amp;theta;).
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/aibook/symbolic systems">symbolic systems</a></span><!--  Array
(
    [slug] =&gt; symbolic systems
    [parent] =&gt; 
    [name] =&gt; symbolic systems
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; Symbolic systems use symbols such as names, places, and concepts whch are manipulated by explicit rules.  In contrast {{subsymbolic systems}}, such as {{neural networks}} use low-level representations such as the weights between nodes.
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/aibook/symmetric matrix">symmetric matrix</a></span><!--  Array
(
    [slug] =&gt; symmetric matrix
    [parent] =&gt; 
    [name] =&gt; symmetric matrix
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; A symmetric matrix is one where the elements either side of the diagonal morror oen another.  That is M&lt;sub &gt;i,j&lt;/sub&gt; = M&lt;sub &gt;j,i&lt;/sub&gt;, or, in other words, M is the matrix  transpose of itself. &lt;br&gt;A 3 x 3 symmetric matrix: &lt;br&gt; &lt;table align=&quot;center&quot;&gt;  &lt;tr&gt;&lt;td align=&quot;center&quot;&gt;3&lt;/td&gt;&lt;td align=&quot;center&quot;&gt;-1&lt;/td&gt;&lt;td align=&quot;center&quot;&gt;2&lt;/td&gt;&lt;/tr&gt;   &lt;tr&gt;&lt;td align=&quot;center&quot;&gt;-1&lt;/td&gt;&lt;td align=&quot;center&quot;&gt;7&lt;/td&gt;&lt;td align=&quot;center&quot;&gt;1&lt;/td&gt;&lt;/tr&gt;   &lt;tr&gt;&lt;td align=&quot;center&quot;&gt;2&lt;/td&gt;&lt;td align=&quot;center&quot;&gt;1&lt;/td&gt;&lt;td align=&quot;center&quot;&gt;-3&lt;/td&gt;&lt;/tr&gt; &lt;/table&gt;
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/aibook/synapse weights">synapse weights</a></span><!--  Array
(
    [slug] =&gt; synapse weights
    [parent] =&gt; 
    [name] =&gt; synapse weights
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; In the brain {{neurones}} connect at {{synapses}}.  This is where signals pass form one to another, but not all connectson are equally strong, the idea of synapse weights cacpures this strength.  In an {{artificial neural network}} these are encoded as a numerical value, but in the brain ithe synapses can physically grow or shrink depending on use, a key part of {{long-term memory}}.
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/aibook/syntactic analysis">syntactic analysis</a></span><!--  Array
(
    [slug] =&gt; syntactic analysis
    [parent] =&gt; 
    [name] =&gt; syntactic analysis
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; Syntactic analysis is the part of {{natural languag eprocessing}} that deals with the structure of the
sentence, that is its {{grammar}}.
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/aibook/syntactic grammar">syntactic grammar</a></span><!--  Array
(
    [slug] =&gt; syntactic grammar
    [parent] =&gt; 
    [name] =&gt; syntactic grammar
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; Syntactic grammars are based solely on the structure of words in sentences, but do not use any knowledge of the meaning of the sentance beyond {{part of speach}} or similar tags, such as whether a wird is a noun or verb.  Syntactic grammars are used in the {{syntactic analysis}} stage of {{natural language processing}} and the output is typically a {{parse tree}}.
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/aibook/synthetic data">synthetic data</a></span><!--  Array
(
    [slug] =&gt; synthetic data
    [parent] =&gt; 
    [name] =&gt; synthetic data
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; Synthetic data is data generated by models or my manipulating real data.  It can be useful for training {{deep neural networks}} when there is insufficient real data or real data is hard to obtain.  For example, 3D simulations can be used to emulate near accident situations, such as pedestraians crossing buy roads or in the dark; these will hopefully be rare in real gatgered video data, but invaluable to help {{autonomous cars}} respond well to exerme events. Synthetic data can also help generalisation, for example by adding noise to images or addings cropped, rotated or resized copies.  Synthetic data has also be sggestd as a way to reduce {{bias}} in data sets.
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/aibook/system 1">System 1</a></span><!--  Array
(
    [slug] =&gt; system 1
    [parent] =&gt; 
    [name] =&gt; System 1
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; In Daniel Kahneman's &quot;Thinking, Fast and Slow&quot;, System 1 refers to the rapid unconscious processes that allow instant reactions, for example, if you start to slip, or you reacxh to catch an approaching ball.  It is on contrats to System 2, which is slower more reasoned thinking.  System 1 is rapid, but may be hard to explain to others. In AI, {{neural networks}} an d{{subsymbolic AI}} in general correspond closely to human System 1 thinking.
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/aibook/system 2">System 2</a></span><!--  Array
(
    [slug] =&gt; system 2
    [parent] =&gt; 
    [name] =&gt; System 2
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; In Daniel Kahneman's &quot;Thinking, Fast and Slow&quot;, System 2 refers to reasoned, conscious thought process, for example weighingnup options for a holiday or deciding what kind of ball to pitch in a game before actually throwing it.  It is on contrast to System 1, which is the faster more unconscious reactions.  System 2 is relatively slow, but open to explanation, communication and discussion. In AI, traditional {{symbolic AI}} in corresponds most closely to human System 2 thinking.
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/aibook/system event">system event</a></span><!--  Array
(
    [slug] =&gt; system event
    [parent] =&gt; 
    [name] =&gt; system event
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; A system event is when the {{software agent}} initiates some form of action based on an internal change, such as completing a computation, or a message from another {{agent}}.  This is in contrast to a {{user event}}, {{timed event}} or {{status change event}}.   In a {{user interface}} the term system event may refer to any event or action generated by the compter system for example an {{agent}} replying to a message or an {{expert system}} requesting information from the user. 
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/aibook/systems response">systems response</a></span><!--  Array
(
    [slug] =&gt; systems response
    [parent] =&gt; 
    [name] =&gt; systems response
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; The 'systems response' is a counter-argument to {{Searle's Chinese Room}}.  It argues that while the person in the room does not understand Chinese, the room as &lt;em&gt;system&lt;/em&gt; does.
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/aibook/tacit knowledge">tacit knowledge</a></span><!--  Array
(
    [slug] =&gt; tacit knowledge
    [parent] =&gt; 
    [name] =&gt; tacit knowledge
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; Tacit knowledge refers to things we know how to do or say, but don;t know that we know.  For example, you might walk the same route home everyday, but find it hard to describe to a friend how to go there.  This is important in {{expert systems}} as much of {{expert knowledge}} is tacit, so {{knowledge elicitation}} techniques need to {{externalise}} this.
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/aibook/technological determinism">technological determinism</a></span><!--  Array
(
    [slug] =&gt; technological determinism
    [parent] =&gt; 
    [name] =&gt; technological determinism
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; Technological determinism is the belief or attitude that technological developments are inevitable and so there is no point in challenging or fighting this 'progress'.  While there are certainly many clear examples where technological devlopments have transformed society, also examples where it is clear that the way in which this acts depends on the social, political, economic and geographical context of societies.  For example, there are numerous examples of wheeled toys, and indeed other wheels, in pre-Columbian Central America, but the wheel was not used for transport.
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/aibook/template matching">template matching</a></span><!--  Array
(
    [slug] =&gt; template matching
    [parent] =&gt; 
    [name] =&gt; template matching
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; Template matching is when text or data is matched aganst simple patterns (templates).  This can be appllied in {{natural language processing}} as an alternatuve to full {{parsing}}, especially in chatbots, for example, using a template such as &quot;tell me about X&quot; (using caps for variables).  It can also be used in {{structured data}}, for example matching &quot;&lt;p class=content&quot;&gt;CONTENT&lt;/p&gt;&quot; in a web page, and in {{data detectors}}.  In {{computer vision}} more compelx templates may be used to represent the visual appearance of objects to be matched against parts of an image.
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/aibook/tensorflow">TensorFlow </a></span><!--  Array
(
    [slug] =&gt; tensorflow
    [parent] =&gt; 
    [name] =&gt; TensorFlow 
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; TensorFlow  is an open. source machine learning and library developed by Google originally for internal use and later released publically.
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/aibook/terminal symbols">terminal symbols</a></span><!--  Array
(
    [slug] =&gt; terminal symbols
    [parent] =&gt; 
    [name] =&gt; terminal symbols
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; In a {{grammar}}, a terminal symbol is one that cannot be expanded into further symbols, for example a pronoun (such as 'it' or 'they').  In terms of a {{parse tree}}, {{non-terminal symbols}} usually label nodes wirthin the body of the tree, whereas terminal symbols label the leaves.
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/aibook/text prediction">text prediction</a></span><!--  Array
(
    [slug] =&gt; text prediction
    [parent] =&gt; 
    [name] =&gt; text prediction
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; Text prediction is used to fill in possible following words of phrases given past text.  Simple versions based on {{n-gram}} frequencies can be used to offer typing suggestions or correct misspellings, but more complex predictions, especially those in {{large-language models}} can continue the text, almost indefinitely as found in chatbots such as {{ChatGPT}} and {{Bard}}.
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/aibook/texture">texture</a></span><!--  Array
(
    [slug] =&gt; texture
    [parent] =&gt; 
    [name] =&gt; texture
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; In {{computer vision}} and {{graphics}} texture is fine-scale detail such as grass, skin or fabric weave, as opposed to areas of solid colour.  Some form of low-level processing is usually needed to detect that an area of texture is in some way all similar.  Ths may be performed using some form of statistical model or in the early layers of a {{deep neural network}}.
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/aibook/theory of mind">theory of mind</a></span><!--  Array
(
    [slug] =&gt; theory of mind
    [parent] =&gt; 
    [name] =&gt; theory of mind
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; Theory of mind refers to the ability to 'put oneself in anothers shoes' that is imagining the world from their view point. 
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/aibook/thought experiment">thought experiment</a></span><!--  Array
(
    [slug] =&gt; thought experiment
    [parent] =&gt; 
    [name] =&gt; thought experiment
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; A thought experiemnt is used in {{philosophy}} to help focus ona. problem or issue.  For example, {{Searle's Chinese room}} and {{Zombie Alice}} are used to discuss isues of {{consciousnes}} and the {{trolley prolem}} to explore ethical choices. 
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/aibook/three-dimensional objects">three-dimensional objects</a></span><!--  Array
(
    [slug] =&gt; three-dimensional objects
    [parent] =&gt; 
    [name] =&gt; three-dimensional objects
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; In {{computer vision}} special algorithms and processes are needed to detect and identify three-dimensional objects from (usually) 2D images.  {{Waltz's algorithm}} is a classic image analysis method for single images following {{edge detection}}.
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/aibook/threshold">threshold</a></span><!--  Array
(
    [slug] =&gt; threshold
    [parent] =&gt; 
    [name] =&gt; threshold
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; A threshold is a critical value beyond which some event occurs or values change drastically.  In biological {{neorones}} and {{artificial neurons}} there is often no reaction for small or moderate amounts of activation from other neorons.  However, if the level of input activation goes beyond some threshold, then the neuron will fully fire.
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/aibook/threshold function">threshold function</a></span><!--  Array
(
    [slug] =&gt; threshold function
    [parent] =&gt; 
    [name] =&gt; threshold function
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; A threshold function changes its value sharply when its input reaches a particular value (the threshold).  The simplest is a {{step function}} zero (or &ndash;1) when it input is less than zero and 1 thereafter.  A hard threshold is simple to implement, but can be hard the learn; so {{neural networks}} typcally use a {{non-linear}} {{sigmod function}} instead.
)
  --></li>
  <li ><span class="name">Tim Berners-Lee</span> - See berners-lee, tim <!--  Array
(
    [slug] =&gt; tim berners-lee
    [parent] =&gt; 
    [name] =&gt; Tim Berners-Lee
    [canonical] =&gt; 
    [see] =&gt; berners-lee, tim 
    [description] =&gt; 
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/aibook/time dependence">time dependence</a></span><!--  Array
(
    [slug] =&gt; time dependence
    [parent] =&gt; 
    [name] =&gt; time dependence
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; A system is time dependent, if its behaviour changes over time based on its past history as well as its current external state.  For example, a simple electrical circuit with resistors will always give the same output for the same inputs ({{time independent}} behaviour), but if it includes capacitors it can have oscilatiry behaviour that depends on past inputs (time dependence).  Time dependence is usually are due to some form of hidden state (such as the levels of charge on capacitors in the electrical circuit) or unobserved state.  Many real phenomena, such as human psychology or the stock market exhibit time dependence, for example, if soemone plays a parctical joke your response may well depend on whether you are happy or stressed.  Materials wtih time dependent behaviour, such as collections of nanoparticles or biofilms, can be used in {{reservoir computing}} to allow it to model complex time dependent phenomena.
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/aibook/time domain">time domain</a></span><!--  Array
(
    [slug] =&gt; time domain
    [parent] =&gt; 
    [name] =&gt; time domain
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; When considering data that has a {{time series}}, such as sound level or weather patterns, we can represent it usingits value at successive time points (val(t=0), val(t=1), ...).  This is called a {{time domain}} representation, and is bot the simplest representation and also the form that most daat is initially gathered.  However, gthe time series can be represented in other ways, notabley {{frequency domain}} that treats the data as a sum of {{sine waves}} or similar functions.
)
  --></li>
  <li ><span class="name"><a href="https://alandix.com/glossary/aibook/timed event">timed event</a></span><!--  Array
(
    [slug] =&gt; timed event
    [parent] =&gt; 
    [name] =&gt; timed event
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; 
)
  --></li>
  <li ><span class="name">TMS</span> - See <span class="has_content"><a href="https://alandix.com/glossary/aibook/truth maintenance system">truth maintenance system</a></span><!--  Array
(
    [slug] =&gt; tms
    [parent] =&gt; 
    [name] =&gt; TMS
    [canonical] =&gt; 
    [see] =&gt; truth maintenance system
    [description] =&gt; 
)
  --></li>
  <li ><span class="name"><a href="https://alandix.com/glossary/aibook/top-down reasoning">top-down reasoning</a></span><!--  Array
(
    [slug] =&gt; top-down reasoning
    [parent] =&gt; 
    [name] =&gt; top-down reasoning
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; 
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/aibook/towers of hanoi">Towers of Hanoi</a></span><!--  Array
(
    [slug] =&gt; towers of hanoi
    [parent] =&gt; 
    [name] =&gt; Towers of Hanoi
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; The Towers of Hanoi is a puzzle where the goal is to move rings between three towesr where smaller rings always have to lie upon larger ones.  
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/aibook/transfer learning">transfer learning</a></span><!--  Array
(
    [slug] =&gt; transfer learning
    [parent] =&gt; 
    [name] =&gt; transfer learning
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; Transfer learning is when a {{machine learning}} system that has been trained on one kind of data, can also be modified or easily retrained to operate on other kinds of data.  For example, a {{deep neural network}} trained on English text may learn deep grammatical patrerns as well as the precise words, and so be able to be modified, with less re-training, to operate on other Germanic languages.  Similarly, game playing systems may be abe to quickly pick up similar games as they have learnt aspects of deep strategy.
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/aibook/transformational analogy">transformational analogy</a></span><!--  Array
(
    [slug] =&gt; transformational analogy
    [parent] =&gt; 
    [name] =&gt; transformational analogy
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; A transformational analogy is a form of {{reasoning}} where one finds a situation that is an {{analogy}} of the current problem, andf then attemtps to transform the solution of the analogus problem to find a sokutikn. to the current one.  For example, when faced with thenproblem of smplifying x&lt;sup&gt;3&lt;/sup&gt; &amp;times; x&lt;sup&gt;5&lt;/sup&gt;, one might use an analogy with 3x + 5x which you know simplfes ot 8x.  Using the analogy that  in algebra (multiplication.power) behaves like (adding,multiplying by numbers), you might then reason that  x&lt;sup&gt;3&lt;/sup&gt; &amp;times; x&lt;sup&gt;5&lt;/sup&gt; = x&lt;sup&gt;8&lt;/sup&gt;.
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/aibook/transformer model">transformer model</a></span><!--  Array
(
    [slug] =&gt; transformer model
    [parent] =&gt; 
    [name] =&gt; transformer model
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; A transformer model operates on {{time series}} or {{sequential data}} using a form of {{attention}}, identifiying past states/tokens that are particualrly related to the current token and then using these in particular as part of predicting future tokens.   This is like the human ability to hear a sentance such as &quot;The cat sat in the deep orange glow of sunset and licked its fur&quot; -- when you read the word 'licked', your mind instantly pulls out the word 'cat' as related and uses that to make sense of the current pont in the sentance.  We use rich {{semantic}} structures to perform this, but transformer models use a {{vector simiilarity}} between a 'key' and 'query' for each token, where the key models the kind of thing that the input token is,  and the query the kind of thing it would like to connect with.
)
  --></li>
  <li ><span class="name"><a href="https://alandix.com/glossary/aibook/transition network">transition network</a></span>- has subentries<ul class="subentries"><li ><span class="name">augmented</span> - See <span class="has_content"><a href="https://alandix.com/glossary/aibook/augmented transition network">augmented transition network</a></span><!--  Array
(
    [slug] =&gt; transition network!augmented
    [parent] =&gt; transition network
    [name] =&gt; augmented
    [canonical] =&gt; 
    [see] =&gt; augmented transition network
    [description] =&gt; 
)
  --></li></ul><!--  Array
(
    [slug] =&gt; transition network
    [parent] =&gt; 
    [name] =&gt; transition network
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; 
    [subentries] =&gt; Array
        (
            [transition network!augmented] =&gt; Array
                (
                    [slug] =&gt; transition network!augmented
                    [parent] =&gt; transition network
                    [name] =&gt; augmented
                    [canonical] =&gt; 
                    [see] =&gt; augmented transition network
                    [description] =&gt; 
                )

        )

)
  --></li>
  <li ><span class="name"><a href="https://alandix.com/glossary/aibook/transparency">transparency</a></span><!--  Array
(
    [slug] =&gt; transparency
    [parent] =&gt; 
    [name] =&gt; transparency
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; 
)
  --></li>
  <li ><span class="name"><a href="https://alandix.com/glossary/aibook/transparent">transparent</a></span><!--  Array
(
    [slug] =&gt; transparent
    [parent] =&gt; 
    [name] =&gt; transparent
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; 
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/aibook/transpose">transpose</a></span><!--  Array
(
    [slug] =&gt; transpose
    [parent] =&gt; 
    [name] =&gt; transpose
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; The transpose of a {{matrix}} M (often written M&lt;sup&gt;T&lt;/sup&gt;) is the matrix reflected along its diagonal. That is given a matrix M&lt;sub&gt;jk&lt;/sub&gt;, its transpose is M&lt;sup&gt;T&lt;/sup&gt;&lt;sub&gt;st&lt;/sub&gt; = M&lt;sub&gt;ts&lt;/sub&gt;.   For example. &lt;br&gt;&lt;br&gt;&lt;table align=&quot;center&quot;&gt; &lt;tr&gt;&lt;th colspan=3 align=center&gt;matrix M&lt;/th&gt;&lt;th&gt;&amp;nbsp;&amp;nbsp;&amp;nbsp;&lt;/th&gt;&lt;th colspan=3 align=center&gt;transpose M&lt;sup&gt;T&lt;/sup&gt;&lt;/th&gt;&lt;/tr&gt; &lt;tr&gt;&lt;td&gt;1&lt;/td&gt;&lt;td&gt;2&lt;/td&gt;&lt;td&gt;3&lt;/td&gt;&lt;td&gt;&lt;td&gt;&lt;td&gt;1&lt;/td&gt;&lt;td&gt;4&lt;/td&gt;&lt;td&gt;7&lt;/td&gt;&lt;/tr&gt;  &lt;tr&gt;&lt;td&gt;4&lt;/td&gt;&lt;td&gt;5&lt;/td&gt;&lt;td&gt;6&lt;/td&gt;&lt;td&gt;&lt;td&gt;&lt;td&gt;2&lt;/td&gt;&lt;td&gt;5&lt;/td&gt;&lt;td&gt;8&lt;/td&gt;&lt;/tr&gt; &lt;tr&gt;&lt;td&gt;7&lt;/td&gt;&lt;td&gt;8&lt;/td&gt;&lt;td&gt;9&lt;/td&gt;&lt;td&gt;&lt;td&gt;&lt;td&gt;3&lt;/td&gt;&lt;td&gt;6&lt;/td&gt;&lt;td&gt;9&lt;/td&gt;&lt;/tr&gt; &lt;/table&gt;
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/aibook/travelling salesman problem">travelling salesman problem</a></span><!--  Array
(
    [slug] =&gt; travelling salesman problem
    [parent] =&gt; 
    [name] =&gt; travelling salesman problem
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; The travelling salesman problem is a common problem with both practical applicatiosn and used as a challenge for optimisation techniques.  A salesperson needs to visit certain locations that are linked by roads or railways, with different travelling costs or times associated wut them.  The task is to plan a route that visits each taregt location, but with the minimum cost for the travel.  The full solution is known to be {{NP complete}}, that is effectively impossible, but various {{statistical techniques}} and {{machine learning}} may achieve near optimal performance in most real cases.
)
  --></li>
  <li ><span class="name">trend adjustments</span> - See <span class="has_content"><a href="https://alandix.com/glossary/aibook/trend removal">trend removal</a></span><!--  Array
(
    [slug] =&gt; trend adjustments
    [parent] =&gt; 
    [name] =&gt; trend adjustments
    [canonical] =&gt; 
    [see] =&gt; trend removal
    [description] =&gt; 
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/aibook/trend removal">trend removal</a></span><!--  Array
(
    [slug] =&gt; trend removal
    [parent] =&gt; 
    [name] =&gt; trend removal
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; Trend removal (or trend adjustments) try to separate long terms trends from shorter term fluctuations in data.  Typically the long-term trend is estimated, often as a linear growth or decay and then subtracted from the data values.  The term is usually applied to {{time series data}}, but can be appled equally to other forms of {{sequential data}}.  Some transformations implicitly remove trend.  For example taking successive differences of a time series (t&lt;sub&gt;i+1&lt;/sub&gt;&amp;minus;t&lt;sub&gt;i&lt;/sub&gt;) effectively removes any linear trend.  Simlarly, taking successive ratios  (t&lt;sub&gt;i+1&lt;/sub&gt;/t&lt;sub&gt;i&lt;/sub&gt;) removes {{exponential}} trend.  
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/aibook/trigger">trigger</a></span><!--  Array
(
    [slug] =&gt; trigger
    [parent] =&gt; 
    [name] =&gt; trigger
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; In an {{agent}} or other software system, an {[event}} can trigger behaviour such as a {{method}} being executed or a {{message}} being sent.  The term trigger is sometimes used as a noun to refer to the condition that gives rise to the action (e.g. &quot;the trigger for turning on the hearing is when the temperature is below 5 degrees&quot;&quot;)  and sometimes as a verb for the invoking of the action (e.g. &quot;the temperature drop triggered the hearting system to start&quot;).
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/aibook/triplestores">triplestores</a></span><!--  Array
(
    [slug] =&gt; triplestores
    [parent] =&gt; 
    [name] =&gt; triplestores
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; Triplestores are a form of {{database}} used to store {{semantic web}} data.. {{RDF}} is encoded as {{subject-predicate-object triples}} and the triplestore stores and indexes these triples in order to make them easy to query and process.  Often {{SPARQL}} is used to query a triplestore.  The {{quadstore}} is an extension of the triplestore to store multiple graphs, that is collections of triples with a common origin or use.
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/aibook/trolley problem">trolley problem</a></span><!--  Array
(
    [slug] =&gt; trolley problem
    [parent] =&gt; 
    [name] =&gt; trolley problem
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; The trolley problem is a thought experiment in {{moral philosophy}}, but has begun to have practical consequences as {{autonomous vehicles}} become common.  Note that the word 'trolley' in the title is used to descibe a small train in some countries.  The trrain is imagiend to be out of control and about to hit a group of (say five) people.  However, tere is a set of pounts that you can control which will send the train down a sidetrack avoiding the people. Unfortunately there is another person on that track.  You have to decide whether to sacrifice this person to save the greater number on the mainline.
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/aibook/truth maintenance system">truth maintenance system</a></span><!--  Array
(
    [slug] =&gt; truth maintenance system
    [parent] =&gt; 
    [name] =&gt; truth maintenance system
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; A truth maintenance system (TMS) is a form of {{non-monotonic reasoning}}.  A TMS links potential assertions (such as 'it is cold' and 'it is winter') and attempts to maintain a coherent set of beliefs about theor truth.  Some are given as facts, others may have their truth determined by assuming defaults where facts are unknown (for example if it is winter one assumes it is also cold).  As new facts arise some of the previous truth values may need to change, for example a temperature reading may show it is avctually an unusually warm wonter's day and hence the truth assigned to 'it is cold' changes. 
)
  --></li>
  <li ><span class="name"><a href="https://alandix.com/glossary/aibook/truth values">truth values</a></span><!--  Array
(
    [slug] =&gt; truth values
    [parent] =&gt; 
    [name] =&gt; truth values
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; 
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/aibook/turing test">Turing test</a></span><!--  Array
(
    [slug] =&gt; turing test
    [parent] =&gt; 
    [name] =&gt; Turing test
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; The Turing test, originally called the Imitation game, was a hypothetical test envisaged by {{Alan Turing}} in 1950 well before computers had the capability of even rudimentary artificial intelligent.  Turing imagined talking to either a computer or a human being at the other end of a text connection.  If it is not possible to tell which is which, then the computer passes the test and can be deemed to be intelligent.  There is debate about whether this really does constitute intelligence, especially now as {{chatbots}} based on {{large language modles}} such as {{chatGPT}} seem to be able to pass the test.
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/aibook/turing, alan">Turing, Alan</a></span><!--  Array
(
    [slug] =&gt; turing, alan
    [parent] =&gt; 
    [name] =&gt; Turing, Alan
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; Alan Turing was a mathematician.  He is famous in part for his work on cryptography during the Second World War including designing the {{Bombe}} to help decrypt {{Enigma machine}} messages.  He was also a theoretician about the principles of computation leading to both the concept of the {{Turing machine}}, a form of universal computation, and positing of the 'Imitation game', otherwise known as the {{Turing test}} to ascertain whether a machine is truly intelligent.
)
  --></li>
  <li ><span class="name"><a href="https://alandix.com/glossary/aibook/uncertain reasoning">uncertain reasoning</a></span><!--  Array
(
    [slug] =&gt; uncertain reasoning
    [parent] =&gt; 
    [name] =&gt; uncertain reasoning
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; 
)
  --></li>
  <li ><span class="name"><a href="https://alandix.com/glossary/aibook/uncertainty">uncertainty</a></span><!--  Array
(
    [slug] =&gt; uncertainty
    [parent] =&gt; 
    [name] =&gt; uncertainty
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; 
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/aibook/unconscious bias">unconscious bias</a></span><!--  Array
(
    [slug] =&gt; unconscious bias
    [parent] =&gt; 
    [name] =&gt; unconscious bias
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; Unconscious bias rerefrs to the way that as humans we make all sorts of assumptions about people based on class, ethnicity, gender or other factors such as how they speak or where they studied.  This is opposed to more conscious {{bias}}, such as explciitly not liking people from a certain country or town based on historic enmity.  Unconscious bias may be based on cultural attitudes that we have implicitly learnt (from media, family or friends) or personal experiences (for example, having had a bad experience with someone of a partcular religion).  The boundary between learning and bias is thus often unclear, but we make choices as individuals and socities as to which are acceptable (e.g. refusing to serve someone acting agressivley) or not (e.g. doing the smae based on the coluor of their skin).  When a {{machine learnng}} system is trisned using pfrevious human decisons or behaviours, it may learn from human {{bias}}, however this is not the only source of  {{bias}} in computer systems.
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/aibook/underdetermined">underdetermined</a></span><!--  Array
(
    [slug] =&gt; underdetermined
    [parent] =&gt; 
    [name] =&gt; underdetermined
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; The weights or parameters in a {{machine learning}} system are underdetermined if there are many different configurations of parameters that all explain the {{training data}}.  For example, if the system is trying to learn simple arithmatic operations, given an input: (2,2) with expected output (4); both 2+2=4, and 2x2=4; so the operation is underdetermined.  While this is obvious in this small case, the {{hidden layers}} in a {{neural network}} often have millions of weights.  Furthermore, of the layers are {{fully connected}}, then any reordering of the internal nodes and weights would geberate equivalent outputs.  This can also occur in {{statistical methods}} such as {{multilinear regression}} is there are too many {{features}}, any of which can contribute.&lt;br&gt;&lt;br&gt;Underdetermined models may {{overlearn}} from the training data leading to poor {{generalisation}} or {{bias}}.  Typically also they can lead to longer training times as the learning algorithm needs to 'decide' between many competing, but equally good, explanations of the data, and so effectively shifts back and forth between them. 
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/aibook/uniform distribution">uniform distribution</a></span><!--  Array
(
    [slug] =&gt; uniform distribution
    [parent] =&gt; 
    [name] =&gt; uniform distribution
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; A uniform distribution is a {{probability distribution}} where the probabilty of each outcome is equal.
)
  --></li>
  <li ><span class="name">uniform resource identifiers</span> - See <span class="has_content"><a href="https://alandix.com/glossary/aibook/uri">URI</a></span><!--  Array
(
    [slug] =&gt; uniform resource identifiers
    [parent] =&gt; 
    [name] =&gt; uniform resource identifiers
    [canonical] =&gt; 
    [see] =&gt; uri
    [description] =&gt; 
)
  --></li>
  <li ><span class="name">uniform resource locators</span> - See <span class="has_content"><a href="https://alandix.com/glossary/aibook/url">URL</a></span><!--  Array
(
    [slug] =&gt; uniform resource locators
    [parent] =&gt; 
    [name] =&gt; uniform resource locators
    [canonical] =&gt; 
    [see] =&gt; url
    [description] =&gt; 
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/aibook/uniform sampling rate">uniform sampling rate</a></span><!--  Array
(
    [slug] =&gt; uniform sampling rate
    [parent] =&gt; 
    [name] =&gt; uniform sampling rate
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; A uniform sampling rate is when a {{time series}} is sampled at fixed intervals, say once a second.  The majority of {{time series anaysis}} techniques assume a uniform sampling rate. However, there are many knds of data where this is not the case, for example, medical data, such as blood results, only occur when a patient visits a doctor or while in hospital the doctir decides a further test is needed; thus they are thus both {{sporadic}} and almost cedrtainly at times that are not typical of times when no sample is taken.  Even where the intention is to have a uniform sampling rate, sensor failures or lost data may mean there are {{missing values}}.
)
  --></li>
  <li ><span class="name"><a href="https://alandix.com/glossary/aibook/unintended bias">unintended bias</a></span><!--  Array
(
    [slug] =&gt; unintended bias
    [parent] =&gt; 
    [name] =&gt; unintended bias
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; 
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/aibook/unique identifier">unique identifier</a></span><!--  Array
(
    [slug] =&gt; unique identifier
    [parent] =&gt; 
    [name] =&gt; unique identifier
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; In a {{database}} or other {{infoirnation system}}, a unique identifier for a data item or record is a combination of features of the data or a generated identifier that is not shared by any other item in the dataset.   For example, in a geographic information system, we might use the physical location of a town as a identifier rather than name, as there are often many towns with the same name.  Often these feature-based identifications are difficult to collect or maintain, or don't even exist, hence the frequent use of generated identifiers such as health or tax numers for people.  Idetifiers may be unique in a single database (e.g. record number) or may be a {{global unique idientifer}} ({GUID}} such as {{URI}}s for the web.
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/aibook/unsupervised learning">unsupervised learning</a></span><!--  Array
(
    [slug] =&gt; unsupervised learning
    [parent] =&gt; 
    [name] =&gt; unsupervised learning
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; In unsupervised learning the {{training set}} is {{unlabelled}}, that is it has no markers to indicate a desired grouping or value.  Unsupervised learning algorithms look for commonalities to group or infer other properties of the data.  This may be used on its own, or be used as the first stage before human labelling or for a subsequent {{supervised learning}} method.
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/aibook/upper bound">upper bound</a></span><!--  Array
(
    [slug] =&gt; upper bound
    [parent] =&gt; 
    [name] =&gt; upper bound
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; An upper bound is an estimate, {{heuristic}} or calculated value that you know lies at or above the largest possible value of an unknown value or function. For example, before you roll a pair of dice, you know that the &lt;em&gt;upper bound&lt;/em&gt; of the sum of the dice scores will be 36 (both dice show six dots).  Knowing an upper bound can sometimes allow whole branches of a {{search tree}} or similar structure to be {{pruned}}.  For example, if you are looking for the largest value in a tree and have found a node with value 7, and then consider an alterantive branch in the tree where a {{heuristic}} says the {{upper bound}} of all items below is 5, you can safely ignore the whole sub-tree. &lt;br&gt;&lt;br&gt;. the opposite of an upper bound is a {{lower bound}}.
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/aibook/uri">URI</a></span><!--  Array
(
    [slug] =&gt; uri
    [parent] =&gt; 
    [name] =&gt; URI
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; A URI (uniform resource identifier) is an extension of the URL.  Every URL is a URI, but not every URI can be accessed in the way a URL can.  It has the same format as URL, but can also refer to non-information entities such as people or places.  The URI acts as a way of refering to the entity so that different information repositories or databases know they are refering to the same thing.  For example https://alandix.com/uri/aibook might be used as a way of talking about the paper book.  This is used in particular in {{RDF}} and other {{semantic web}} techologies.
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/aibook/url">URL</a></span><!--  Array
(
    [slug] =&gt; url
    [parent] =&gt; 
    [name] =&gt; URL
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; A URL (uniform resource locator) is the familar address in the top bar of a {{web browser}}.  For example, https://alandix.com/glossary/aibook is the URL of this glossary.Unlike earlier {{hypertext}} a URL is both globally unique  and can be accessed to give the resource (most commonly a web page or image).
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/aibook/user event">user event</a></span><!--  Array
(
    [slug] =&gt; user event
    [parent] =&gt; 
    [name] =&gt; user event
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; A user event in a {{user interface}} or when a human communicates with an {{agent}}, is when the user initiates some form of action or message, for example asking a question to an {{agent}} or clicking a mouse in a {{user interface}}.  This is in contrast to a {{system event}}.
)
  --></li>
  <li ><span class="name"><a href="https://alandix.com/glossary/aibook/user interface">user interface</a></span><!--  Array
(
    [slug] =&gt; user interface
    [parent] =&gt; 
    [name] =&gt; user interface
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; 
)
  --></li>
  <li ><span class="name"><a href="https://alandix.com/glossary/aibook/validation rules">validation rules</a></span><!--  Array
(
    [slug] =&gt; validation rules
    [parent] =&gt; 
    [name] =&gt; validation rules
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; 
)
  --></li>
  <li ><span class="name"><a href="https://alandix.com/glossary/aibook/value network">value network</a></span><!--  Array
(
    [slug] =&gt; value network
    [parent] =&gt; 
    [name] =&gt; value network
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; 
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/aibook/variable-order markov models">variable-order markov models</a></span><!--  Array
(
    [slug] =&gt; variable-order markov models
    [parent] =&gt; 
    [name] =&gt; variable-order markov models
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; While a simple {{Markov model}}  only uses the last state/token to predict the next one, a variable-order Markov model may use several past states.  For example a third-order Markov model, gives the probabilty of the state S&lt;sub&gt;t&lt;/sub&gt; as a function of the last three states: &lt;br&gt; &amp;nbsp; &amp;nbsp; &amp;nbsp; P(S&lt;sub&gt;t&lt;/sub&gt;=s) &amp;nbsp;  = &amp;nbsp;  f(s&lt;sub&gt;t-1&lt;/sub&gt;,s&lt;sub&gt;t-2&lt;/sub&gt;,s&lt;sub&gt;t-3&lt;/sub&gt;)
)
  --></li>
  <li ><span class="name"><a href="https://alandix.com/glossary/aibook/variance">variance</a></span><!--  Array
(
    [slug] =&gt; variance
    [parent] =&gt; 
    [name] =&gt; variance
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; 
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/aibook/vcard ontology">vCard ontology</a></span><!--  Array
(
    [slug] =&gt; vcard ontology
    [parent] =&gt; 
    [name] =&gt; vCard ontology
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; The vCard ontology is a standard way of describing contact details for people ... that is the information one might find in a 'visiting card'.  It arose as an {{emergent standard}} for interchange between contact management systems and as a web {{microformat}}.  The {{RDF}} form is used extensively in {{semantic web}} {{ontologies}}.
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/aibook/vergence">vergence</a></span><!--  Array
(
    [slug] =&gt; vergence
    [parent] =&gt; 
    [name] =&gt; vergence
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; In a {{stereo camera}}, the vergence is the horizontal angle between two cameras in the head, which allows both cameras to fixate on one object and also can be used to gauge the object's distance.
)
  --></li>
  <li ><span class="name"><a href="https://alandix.com/glossary/aibook/version-space algorithm">version-space algorithm</a></span><!--  Array
(
    [slug] =&gt; version-space algorithm
    [parent] =&gt; 
    [name] =&gt; version-space algorithm
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; 
)
  --></li>
  <li ><span class="name">Vilfredo Pareto</span> - See <span ><a href="https://alandix.com/glossary/aibook/pareto, vilfredo">Pareto, Vilfredo</a></span><!--  Array
(
    [slug] =&gt; vilfredo pareto
    [parent] =&gt; 
    [name] =&gt; Vilfredo Pareto
    [canonical] =&gt; 
    [see] =&gt; pareto, vilfredo
    [description] =&gt; 
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/aibook/visual analytics">visual analytics</a></span><!--  Array
(
    [slug] =&gt; visual analytics
    [parent] =&gt; 
    [name] =&gt; visual analytics
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; Visual analytics combines {{visualisation}}, {{machine learning}} and {{statistical analysis}} in an interactuve system to aid human {{sense making}} and {{decision making}} especially when delaing with complex data.
)
  --></li>
  <li ><span class="name"><a href="https://alandix.com/glossary/aibook/visualisation">visualisation</a></span><!--  Array
(
    [slug] =&gt; visualisation
    [parent] =&gt; 
    [name] =&gt; visualisation
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; 
)
  --></li>
  <li ><span class="name">visualising data</span> - See <span ><a href="https://alandix.com/glossary/aibook/visualisation">visualisation</a></span><!--  Array
(
    [slug] =&gt; visualising data
    [parent] =&gt; 
    [name] =&gt; visualising data
    [canonical] =&gt; 
    [see] =&gt; visualisation
    [description] =&gt; 
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/aibook/voxel">voxel</a></span><!--  Array
(
    [slug] =&gt; voxel
    [parent] =&gt; 
    [name] =&gt; voxel
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; A voxel is the 3D equivalent of a {{pixel}}, that is a small rectangular (usually cubic) patch of space.  For example, if a room is 5 metres long, by 4 metres wide, by 3metres high (5mx4mx3m), we might divide it into 5000x4000x3000 1mm cubes for the purposes of indexing objects or performing 3D rendering.  While photographs are usually 2D, various forms of medical scanners, such as ultrasound or MRI generate 3D views of the patient.  However, these may have non-cubic voxels as they create a series of frames or slices and the gap between slices (usually deterimined by mechanical movement of the device or patient) may be different from the resolution within a slice.
)
  --></li>
  <li ><span class="name"><a href="https://alandix.com/glossary/aibook/waltz's algorithm">Waltz's algorithm</a></span><!--  Array
(
    [slug] =&gt; waltz's algorithm
    [parent] =&gt; 
    [name] =&gt; Waltz's algorithm
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; 
)
  --></li>
  <li ><span class="name">Watson (IBM)</span> - See <span class="has_content"><a href="https://alandix.com/glossary/aibook/ibm watson">IBM Watson</a></span><!--  Array
(
    [slug] =&gt; watson
    [parent] =&gt; 
    [name] =&gt; Watson (IBM)
    [canonical] =&gt; 
    [see] =&gt; ibm watson
    [description] =&gt; 
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/aibook/wavelength">wavelength</a></span><!--  Array
(
    [slug] =&gt; wavelength
    [parent] =&gt; 
    [name] =&gt; wavelength
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; The wavelength is the peak-to-peak distance of a wave.  For sea waves this may vary form a few centimetres for ripple to may metres for deep sea waves.  For sounds this varies from a few centimetres for notes at the top of a piano to several metres for those at the bottom.  For light the wavelength is between around 400nm (violet) to 700nm (red).   This is to be contrasted with the {{frequency}}, which os the time taken for the wave to move for. peak-to-peak at a fixed location.  For sound waves the two are relatred by the speed of sound, for electriomagnetic waves the speed of lught: &lt;br&gt; &amp;nbsp; &amp;nbsp; &amp;nbsp; wavelength / frequency &amp;nbsp;  =  &amp;nbsp; speed
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/aibook/wavelet">wavelet</a></span><!--  Array
(
    [slug] =&gt; wavelet
    [parent] =&gt; 
    [name] =&gt; wavelet
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; A wavelet is a small function, which has finite/local extent in time or space.  For example: &lt;br&gt; &amp;nbsp; &amp;nbsp; &amp;nbsp;  f&lt;sub&gt;s(t)  &amp;nbsp;= &amp;nbsp; 1 - |t-s|  &amp;nbsp; if (t-s) &lt; 1 ,  &amp;nbsp; =  &amp;nbsp;0 otherwise &lt;b&gt;&lt;br&gt; Wavletes can be one diennsional, for use over {{time series}} series or {{sequential data}}, 2D for images, or higher dimensions.   Wavelets usually form part fo a collection with different wavelests at different sizes and locations, so that any \term{time series} or image can be expressed as a sum of the wavekets in the collection -- called a {{wavelet transform}} of the function.
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/aibook/wavelet transform">wavelet transform</a></span><!--  Array
(
    [slug] =&gt; wavelet transform
    [parent] =&gt; 
    [name] =&gt; wavelet transform
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; A wavelet transform decomposes a function or time series into a sum of products of {{wavelets}} from a collection.  The collection is often defined mathematically using a {{kernal}}, which then mans thet thay have good properties.  It is rather like a {{Fourier transform}}, as the wavelets typically opertae at different tmeoral or spatial scals, but whereas the sine/cos funcitons used in the {{Fourier transform}} extend over the etire temproal or spatial domain, wavelets are local.
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/aibook/weak ai">weak AI</a></span><!--  Array
(
    [slug] =&gt; weak ai
    [parent] =&gt; 
    [name] =&gt; weak AI
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; Proponents of weak AI take the position that it is sufficent if AI creates computer systems that perform in an intelligent manner even if they do so in very diffeernt ways to humans.  That is they are not trying to deeply emulate human intelligence in the way that proponents of {{strong AI}} aim to do.
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/aibook/weak learners">weak learners</a></span><!--  Array
(
    [slug] =&gt; weak learners
    [parent] =&gt; 
    [name] =&gt; weak learners
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; In {{machine learniong}} weak learners are sub-models that are in themselves not that good, but can potentially be combined, or modified in someway to make them stringer.  For example, {{random forest}} techniques create large numbers of {[decision trees}} each of which may not be that effective, but when combined by higher-level decision algorithm (such as weighted voting), have proven very powerful.  
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/aibook/web of data">web of data</a></span><!--  Array
(
    [slug] =&gt; web of data
    [parent] =&gt; 
    [name] =&gt; web of data
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; The {{world wide web}} that we browse consists of human rewadable web paes.  However, the web was originallyconceived in {{CERN}} as a way to share data.  {{Tim Berners-Lee}}'s vision is for a &lt;em&gt;web of data&lt;/em&gt; that is machine readable, so that automated systems and reasoning can apply at a global scale. Central to this vision is the {{semantic web}}, which means that not only the data itself, but also the {{meta-information}} necessary to interpret it are available in a machine readable and linkable fashion.
)
  --></li>
  <li ><span class="name">web ontology language</span> - See <span class="has_content"><a href="https://alandix.com/glossary/aibook/owl">OWL</a></span><!--  Array
(
    [slug] =&gt; web ontology language
    [parent] =&gt; 
    [name] =&gt; web ontology language
    [canonical] =&gt; 
    [see] =&gt; owl
    [description] =&gt; 
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/aibook/web science">web science</a></span><!--  Array
(
    [slug] =&gt; web science
    [parent] =&gt; 
    [name] =&gt; web science
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; Web Science is a cross-disciplinary field studying the web as a wholistic entity including humans and machines.  The web could simply be seen as a graph of interconnected pages and other data.  Web science recognises that this data of the web is part of a larger system that also includes people interacting with and through the web, and physiclal processes being monitored to generate web data and also being controlled through the web.  Web science thefore interacts with many fields including {{data science}}, {{human computer interaction}} and the {{internet of things}}
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/aibook/web scraping">web scraping</a></span><!--  Array
(
    [slug] =&gt; web scraping
    [parent] =&gt; 
    [name] =&gt; web scraping
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; Web scraping involves taking web pages intended for human clnsumption and extracting key data from it, using the structure of the page for some form of {{template matching}} or similar process.  For example, some of the data elemints on the page may have specific CSS classes that can be used to work out where they are and then extract their data.  The output of web scraping is data that can then be used for other processing algorithms.
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/aibook/weizenbaum, joseph">Weizenbaum, Joseph</a></span><!--  Array
(
    [slug] =&gt; weizenbaum, joseph
    [parent] =&gt; 
    [name] =&gt; Weizenbaum, Joseph
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; Joseph Weizenbaum was an early researcher and theorist in AI and in particular created the chatbot {{ELIZA}}.  Weizenbaum was a sceptic about the ultimate potential for computers to be truly intelligent.
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/aibook/white matter">white matter</a></span><!--  Array
(
    [slug] =&gt; white matter
    [parent] =&gt; 
    [name] =&gt; white matter
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; White matter is the material in the centre of the brain as compated with the {{grey matter}} at the outside.  The white matter that is principally composed of {{axons}}, the nerve fibres that connect the bodies of {{neorones}} that are concentretd in the {{grey matter}}.
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/aibook/windowing">windowing</a></span><!--  Array
(
    [slug] =&gt; windowing
    [parent] =&gt; 
    [name] =&gt; windowing
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; Many techniques for {{time-series}} or {{sequential data}} work by taking a fixed size window, that is the last N items of the series, and then use each wondow as if it were separate data, eiethre for training ro execution.  Examples include {{moving average}} methods in {{time series analysis}}, simple {{Markov models}} and applying {{neural networks}} to the data windows.  Wondowing methods are always {{finite impluse response}} as data outside of the wondow cannot affect the current response.
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/aibook/winner takes all">winner takes all</a></span><!--  Array
(
    [slug] =&gt; winner takes all
    [parent] =&gt; 
    [name] =&gt; winner takes all
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; Many classification or decsiion making algorithms output a set of scores, or weights for each potential outcome rather than a single classification or decision.  A winner takes all strategy takes the highest scrore and uses that to make a single choice.  This is a low risk stregtegy, but often means that systems do not sufficiently explore potential behaviours.   Alternative strategies include using the scores to create a probability and then randomly make the final decision based on that.
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/aibook/winograd, terry">Winograd, Terry</a></span><!--  Array
(
    [slug] =&gt; winograd, terry
    [parent] =&gt; 
    [name] =&gt; Winograd, Terry
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; Terry Winograd is one of the major figures in the development of AI and in particular created  the blocks world {{SHRLDU}}.
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/aibook/wisdom of the crowds">wisdom of the crowds</a></span><!--  Array
(
    [slug] =&gt; wisdom of the crowds
    [parent] =&gt; 
    [name] =&gt; wisdom of the crowds
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; The wisdom of the crowds is the thesis that if you ask sufficient people a question then even if each is individually not ver knowledgable, the consensus view is remarkably accurate.  This seems to work for certain kinds of problem where tno expert knowledge is required, but if, say, the question concerned qunatum string theory, it seems less lilely that a random crowd would be helpful unless it happend to be a crowd of quantum string theorists.  However, a computatuonal variant of this princple does seem to work well in {{ensemble}} methods in AI.
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/aibook/wizard">wizard</a></span><!--  Array
(
    [slug] =&gt; wizard
    [parent] =&gt; 
    [name] =&gt; wizard
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; A wizard is a part of the user interface in a system that helps a user confugure or otherwsie adjust the system usually using simple question-answer or form-filling interactions.  Wizards tend to be pre-programmed for very specific aspects of a system, but there is the potential for intelligent assustants to be far more flexible.
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/aibook/word vector">word vector</a></span><!--  Array
(
    [slug] =&gt; word vector
    [parent] =&gt; 
    [name] =&gt; word vector
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; A word vector is the value associated with a word in {{word2vec}}.  It captures some level of meaning, for example, if WV('dog'), WV('puppy'), WV('cat') and WV('kitten') are the word vectors assocated with 'dog', 'puppy', 'cat' and 'kitten' respectvely, then the vector difference WV('dog')&ndash;WV('puppy') will typically be close to WV('cat')&ndash;WV('kitten').  Tis makes it possible to asnwre questions such as &quot;dog is to puppy as cat is to what?&quot;
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/aibook/word2vec">word2vec</a></span><!--  Array
(
    [slug] =&gt; word2vec
    [parent] =&gt; 
    [name] =&gt; word2vec
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; word2vec creates a vector representation of a word that captures certaib aspects of the menaing of the word.  It works by first creating a 'Skip-gram model' that predicts the likelihood of curroundng words, and then extracts an internal layer of this model to form a {{latent space}}.  The value of a word in this latent space is called a {{word vector}}.  The geometry of the {{latent space}} often captures crucially relationships, so that, for example, the vector difference between 'dog' and  'puppy' is similar to that between 'cat' and 'kitten'.  These word vectors can be used in further steps of an {{NLP}} process.
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/aibook/world wide web">World Wide Web</a></span><!--  Array
(
    [slug] =&gt; world wide web
    [parent] =&gt; 
    [name] =&gt; World Wide Web
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; Although we mostly think of the web in terms of human readable web pages, the invention of the {{World Wide Web}} by {{Tim Berners-Lee}} was originally focused on helping physicists at {{CERN}} to share data.  This vision is expressed in the {{semantic web}} a number of technologies aimed at making the knowledge on the web also understandable by machines.  In addition, AI is being used extensively, for example in {{recommender systems}} that suggest items for you in media and shopping web sites, and to celect items for feeds in social media applications. 
 Furthermore, even human readable pages are being used as data for {{machine learning}} notably {{large language models}}.  Indeed, it is this huge volume of {{big data}} available on the web which has been one of the main drivers for the latest wave of AI success.
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/aibook/xcon">XCON</a></span><!--  Array
(
    [slug] =&gt; xcon
    [parent] =&gt; 
    [name] =&gt; XCON
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; XCON was an early {{expert system}} developed by DigitalElectronics Corporation to configure VAX computer systems
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/aibook/xor problem">XOR problem</a></span><!--  Array
(
    [slug] =&gt; xor problem
    [parent] =&gt; 
    [name] =&gt; XOR problem
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; Early {{single-layer perceptrons}} were not able to solve the XOR problem, that is how to calculate the exclusive or of two inputs.
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/aibook/zero-sum game">zero-sum game</a></span><!--  Array
(
    [slug] =&gt; zero-sum game
    [parent] =&gt; 
    [name] =&gt; zero-sum game
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; This is a game where one player's gain means the other player's loss.   If it is a simple win-lose game, then only one player can win, and eiether the other loses, or they both draw.  In a game with scores, it emasn that the total score is fixed.  {{Noughts and crosses}} is an example of a simple win-lose zero-sum game.  Some {{game playing algorithms}} such as {{alpha-beta pruning}} only work for zero-sum games.    This is the opposite of a {{non-zero-sum game}}.
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/aibook/zombie alice">Zombie Alice</a></span><!--  Array
(
    [slug] =&gt; zombie alice
    [parent] =&gt; 
    [name] =&gt; Zombie Alice
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; Zombie Alice is used in thought experiments by philosophers discussing {{consciousness}}.  They imagine zombie Alice wo is just like a human Alice, in all ways except has no real mind.  If Zombie Alice behaves in all respects like a conscious person  are they indeed conscious, or does this show that consciousness is more than behaviour?
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/aibook/zoom">zoom</a></span><!--  Array
(
    [slug] =&gt; zoom
    [parent] =&gt; 
    [name] =&gt; zoom
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; In a camera, the zoom increases magnification and thus allows distant obecst to be viewed more easily.
)
  --></li>
</ul>
</div>
</body>
</html>