Exploiting decomposability using recombination in genetic algorithms: An exploratory discussion

Abstract

On certain classes of problems, recombination is more effective if the parents that are being recombined share common subsolutions. These common subsolutions can be used to decompose the recombination space into linearly independent subproblems. If a problem can be decomposed into k subproblems, a single greedy recombination can select the best of 2k possible offspring. The idea of exploiting decomposability works well for the Traveling Salesman Problem, and appears to be applicable to other problems such as Graph Coloring. For Search Based Software Engineering, these ideas might be useful, for example, when applying Genetic Programming to fix software bugs in large programs. Another way in which we might achieve decomposability is by exploiting program modularity and reoccurring program patterns. © 2011 Springer-Verlag.