Semantic based crossovers in Tree-Adjoining Grammar Guided Genetic Programming

Abstract

In this paper we propose an approach to implementing new semantic based crossover operators in Tree-Adjoining Grammar Guided Genetic Programming (TAG3P). The design of the new crossover operators is based on the non-fixed arity (also called feasibility) property of TAG-based representation in TAG3P. The new operators are then tested on a family of benchmark symbolic regression problems and compared with standard Genetic programming (GP), GP with Semantic Similarity based Crossover (SSC), and TAG3P. The results show that TAG3P with the new operators significantly outperforms GP, GP with SSC, and TAG3P.