Associative Memory and Mutual Information in a Chaotic Neural Network
Introducing Function Typed Synaptic Weights

Abstract

So far, in associative memory search problems chaotic neural networks have constant synaptic weights to store patterns. In this paper, we propose a chaotic neural network(CNN) which has function typed synaptic weights to store patterns in order to make a better performance of the retrieval of the stored patterns. In stored patterns retrieval simulation, it is clarified that our proposed method is superior to the conventional method, that is, which has constant synaptic weights. Furthermore we propose an algorithm to calculate the mutual information in a CNN and show that the mutual information in the CNN, which are on the edge of chaos, gets the biggest values.