Gate-level Synthesis of Boolean Functions using Information Theory Concepts

In this paper we apply information theory concepts to evolutionary Boolean circuit synthesis. We discuss the schema destruction problem when simple conditional entropy is used as fitness function. The design problem is the synthesis of Boolean functions by using the minimum number of binary multiplexers. We show that the fitness landscape of normalized mutual information exhibits better characteristics for evolutionary search than the landscape of simple mutual information. A comparison of minimum evolved circuits shows the potential of information theory concepts.