A new subgraph crossover for cartesian genetic programming

Abstract

While tree-based Genetic Programming is often used with crossover, Cartesian Genetic Programming is mostly used only with mutation as genetic operator. In this paper, a new crossover technique is introduced which recombines subgraphs of two selected graphs. Experiments on symbolic regression, boolean functions and image operator design problems indicate that the use of the subgraph crossover improves the search performance of Cartesian Genetic Programming. A preliminary comparison to a former proposed crossover technique indicates that the subgraph crossover performs better on our tested problems.