Controlling Tree Size Growth in Genetic Programming

This paper presents an approach to solve the parsimony, or a tree size growth, problem in Genetic Programming (GP). The approach is formulated as a multiobjective optimisation problem where parsimony is included as one of the objectives. This Multi-Objective Genetic Programming (MOGP) method is tested using the 6-Multiplexer benchmark problem. The MOGP is showb to consistently perform better than approaches which include parsimony pressure, a penalty to the programme size, as part of a single objective function. The approach also results in a considerable reduction in a computational processing time as the population evolves toward more parsimonious tree-structured representations