Proposal of Surrogate Model for Genetic Programming Based on Program Structure Similarity

摘要

This paper proposes a novel surrogate model for genetic programming that estimates the fitness of each individual by using the tree structure similarity. In particular, the fitness of each individual is estimated with the nearest neighbor method by comparing each individual with the evaluated population. We conduct an experiment to investigate the effectiveness of the proposed method. In the experiment, we compare genetic programming with and without the proposed surrogate model on the symbolic regression problem. We assess the convergence speed and the discovery ratio of the optimum program. The experimental result reveals that the proposed method improves the convergence speed of genetic programming while maintaining the discovery rate of the optimum program.