Exploring Geometric Semantic Genetic Programming for the evolution of Classifiers [poster]

Abstract

Genetic Programming (GP) is the concept of building and evolving computer programs according to the Darwin's Theory of Evolution. Many different algorithms fall under its umbrella, and this document takes particular interest in one called Geometric Semantic Genetic Programming (GSGP). This specific variation uses an approach focused on the semantics of the evolved computer programs and utilizes genetic operators with specific properties that allow them to induce a uni-modal fitness landscape on any type of problem. The results displayed by this technique are excellent, particular within the realm of solving regression problems. The aim of this work is to explore the potential of GSGP techniques, particularly in the realm of classification problems and aims to develop implementations specifically designed for multi-class problems.