Autoencoding with a Classifier System

Autoencoders enable data dimensionality reduction and are a key component of many learning systems. This article explores the use of a learning classifier system to perform autoencoding. Initial results using a neural network representation in the classifiers suggest that a classifier system can be an effective approach to data reduction. The approach adaptively subdivides the input domain into local approximations---in effect, small autoencoders---that together cover the problem space. By allowing the number of neurons in the autoencoders to evolve, local solutions of different complexity emerge. Self-adaptative mutation tunes the rate of gradient descent in each layer, reducing the parameter optimisation task.