Using instruction matrix based genetic programming to evolve programs

Abstract

In Genetic Programming (GP), evolving tree nodes separately would be an ideal approach to reduce the huge solution space of GP. We use Instruction Matrix based Genetic Programming (IMGP) to evolve tree nodes separately while taking into account their interdependencies in the form of subtrees. IMGP uses an Instruction Matrix (IM) to maintain the statistical data of tree nodes and subtrees. IMGP extracts program trees from IM, and updates IM with the information of the extracted program trees. The experiments have verified that the results of IMGP are better than those the related GP algorithms in terms of the qualities of the solutions and the number of program evaluations. © Springer-Verlag Berlin Heidelberg 2007.