Automatic synthesis of associative memories through genetic programming: A first co-evolutionary approach

Abstract

Associative Memories (AMs) are mathematical structures specially designed to associate input patterns with output patterns within a single stage. Since the last fifty years all reported AMs have been manually designed. The paper describes a Genetic Programming based methodology able to create a process for the automatic synthesis of AMs. It paves a new area of research that permits for the first time to propose new AMs for solving specific problems. In order to test our methodology we study the application of AMs for real value patterns. The results illustrate that it is possible to automatically generate AMs that achieve good recall performance for problems commonly used in pattern recognition research. © 2010 Springer-Verlag Berlin Heidelberg.