On reducing the influence of noise in a new model for optimal linear associative memory

Abstract

The authors propose a new model for linear memory. In this work, a synaptic matrix consists of not only the stored input and output patterns, but also of the injected attached patterns, which are weighted periodic inverse-repeat pseudorandom patterns. When the injected patterns are the stored input patterns, Kohonen's model is obtained. As such, Kohonen's model is a special case of the model proposed here. When the real pattern is contaminated by colored noise, recalling the stored pattern is superior to that obtained from Kohonen's pseudoinverse learning rule. The authors' learning rule can reduce the colored noise influence on the optimal linear associative memory and is shown to be optimal in the least mean square sense. The theoretical results are illustrated with computer simulation. >