Abstract

Associative neural memory of the Hopfield type, where each element (neuron) has a restricted amount of connections is considered as a cellular-automaton-like network. The goal of the investigation is to determine the degradation of memory capacity and efficiency of pattern retrieval caused by the decrease of connection number. As a result, conditions of individual stability and 1-attractivity of stored patterns are obtained being expressed in terms of neighborhood configurations of each neuron. Based on the above conditions a method of weight computation which insures stored patterns stability is introduced. Some results of cellular-neural associative memory simulation are presented.

File
Issue
Pages
1-16