A categorizing associative memory using sparse coding and an adaptive classifier

This paper presents a neural network which stores and retrieves sparse codings categorically, the codings being random realizations of a sequence of biased (0,1) Bernoulli trials. The neural network, denoted by the categorizing associative memory (CAM), consists of two essential functional modules: (1) an adaptive classifier (AC) module which categorizes input data and which bears some resemblance to the ART2a model, and (2) an associative memory module which stores a number of input patterns in each category according to the Hebbian rule, after the AC-module has stabilized its learning of the category.