hyperparameters

Terms from Artificial Intelligence: humans at the heart of algorithms

Page numbers are for draft copy at present; they will be replaced with correct numbers when final book is formatted. Chapter numbers are correct and will not change now.

Hyperparameters are configuration settings for machine learning systems such as the number and size of layers in a deep neural netwrok. These are differentiated from the parameters within the model, such as weights in a neural network that are trained by the machine learning algorithm. Often these hyperparameters are chosen by hand based on experience or rules of thumb; however they may also be derived by higher-level algorithms or meta-level machine learning. Care needs to be taken to ensure that one is not simply choosing hyperparameters that work for the training data, a form of overfitting.