Pattern recognition using a neural network with the short term memory

Abstract

The paper deals with the modification of the Hamming neural network designed for solving the problems of pattern recognition. It is proposed to divide the memory of a neural network in the short term and long term parts. To the Hamming network the additional layers and modulators are added, which provide the property of plasticity-stability of memory, like the networks in the adaptive resonance theory. An algorithm for the short-term memory consolidation is proposed that is based on the frequency of encountering the components of stored images.