Towards Efficient Semantic Mutation in CGP: Enhancing SOMOk

Towards Efficient Semantic Mutation in CGP: Enhancing SOMOk

Stať ve sborníku mimo WoS a Scopus

Genetic Programming (GP) and its variants have proven to be promising techniques for solving problems across various domains. However, GP does not scale well, particularly when applied to symbolic regression in the Boolean domain. To address this limitation, a semantically oriented mutation operator (SOMO) has been proposed and integrated with Cartesian Genetic Programming (CGP). Nevertheless, like standard GP, even SOMO suffers in some cases from bloat - an excessive growth in solution size without a corresponding performance gain. This work introduces SOMOk-TS, an extension of SOMO that incorporates the so-called Tumor Search strategy to identify and preserve reusable substructures. By managing diversity through an immune-inspired mechanism, SOMOk-TS promotes the reuse of substructures, thereby reducing computational overhead. It achieves significantly lower execution times while maintaining or improving solution compactness, highlighting its potential for scalable and efficient evolutionary design.

Genetic Programming (GP) and its variants have proven to be promising techniques for solving problems across various domains. However, GP does not scale well, particularly when applied to symbolic regression in the Boolean domain. To address this limitation, a semantically oriented mutation operator (SOMO) has been proposed and integrated with Cartesian Genetic Programming (CGP). Nevertheless, like standard GP, even SOMO suffers in some cases from bloat - an excessive growth in solution size without a corresponding performance gain. This work introduces SOMOk-TS, an extension of SOMO that incorporates the so-called Tumor Search strategy to identify and preserve reusable substructures. By managing diversity through an immune-inspired mechanism, SOMOk-TS promotes the reuse of substructures, thereby reducing computational overhead. It achieves significantly lower execution times while maintaining or improving solution compactness, highlighting its potential for scalable and efficient evolutionary design.

Genetic Programming, Boolean function learning

Genetic Programming, Boolean function learning

PLEVAČ, L.; VAŠÍČEK, Z.

14.07.2025

Association for Computing Machinery

Malaga

979-8-4007-1464-1

Proceedings of the Genetic and Evolutionary Computation Conference Companion

2172

2176

5