We propose a novel synthesis procedure of an associative memory with genetic algorithm. The associative memory consists with a discrete-time neural network whose connection coefficient takes only two value, that is ±1. In our proposed synthesis procedure, the connection coefficient matrix is regarded as a gene, and the gene is evolved by a genetic algorithm. By using an proposed fitness function that is based on association rate of desired memories from similarly input patterns, the genes so that takes high evaluation value is left. Many of previous related works use auto-correlation matrix to store desired memories. Our synthesis procedure does not use auto-correlation matrix as the connection matrix. but this system generate the connection matrix resembled correlation matrix. In this article. we consider the relationship between the generated matrix and the auto-correlation matrix.
展开▼
