redundancy

Terms from Artificial Intelligence: humans at the heart of algorithms

The glossary is being gradually proof checked, but currently has many typos and misspellings.

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 clearning. However, redundancy can also be critical in fault tolerant systems allowing a level of recovery if a sensor fails or data is lost.
Redundancy may be precise, in that one piece of data can be calculated from others. Alternatively, the redundant data items may be statistically related, for example where several values in some way measure or represent 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 obtain a smaller number of more accurate readings.
In activity recognition systems redundant sensors give more robust detection. It also means that as new sensors are added they can be correlated with existing sensors, adding to redundancy especially if existing sensors fail or are remove, thus allowing deployments of sensors to change whilst retaining higher-level reasoning.

Used in Chap. 19: page 303

Also known as redundant data