Geometric Semantic Grammatical Evolution

Abstract

Geometric Semantic Genetic Programming (GSGP) is a novel form of Genetic Programming (GP), based on a geometric theory of evolutionary algorithms, which directly searches the semantic space of programs. In this chapter, we extend this framework to Grammatical Evolution (GE) and refer to the new method as Geometric Semantic Grammatical Evolution (GSGE). We formally derive new mutation and crossover operators for GE which are guaranteed to see a simple unimodal fitness landscape. This surprising result shows that the GE genotypephenotype mapping does not necessarily imply low genotype-fitness locality. To complement the theory, we present extensive experimental results on three standard domains (Boolean, Arithmetic and Classifier).