Matching Subtrees in Genetic Programming Crossover Operator

Authors: M. Šlapák, Roman Neruda
Publication year: 2017
Publication type: Konferenční příspěvek (mezinárodní konf.)
Periodical: ICNC-FSKD 2017. Proceedings of International Conference on Natural Computation, Fuzzy Systems and Knowledge Discovery
Projects: Automated Knowledge and Plan Modeling for Autonomous Robots
Bibliographic citation: ŠLAPÁK, M., NERUDA, Roman. Matching Subtrees in Genetic Programming Crossover Operator. In: ICNC-FSKD 2017. Proceedings of International Conference on Natural Computation, Fuzzy Systems and Knowledge Discovery. Piscataway: IEEE, 2017, S. 208-213. ISBN 978-1-5386-2165-3
Dostupné z: doi:10.1109/FSKD.2017.8393091
DOI: 10.1109/FSKD.2017.8393091
ID: 491993

Abstract

In this paper we study techniques that should reduce the destructive impact of crossover in genetic programming. The quality of crossover offsprings is often lower than ancestors due to the fact that a small change in individual's genotype tree structure has a great impact to its phenotype. Therefore we propose and test several methods for matching subtrees to find the best possible cutting point for crossover of trees. Our approach utilizes the adaptive probability of operators with the intent to reinforce the well-performing operators. A relation to the semantic genetic programming approach is also investigated. The experimental results show that the average arity based technique performs best from the proposed methods.