An analysis of diversity of constants of Genetic Programming

Abstract

This paper conducts an investigation into the manner in which constants evolve during the course of GP run. It starts by describing an intuitive Gaussian type mutation for constants and showing that its ability to produce small changes in individuals leads to a high performance. It then demonstrates the surprising result that, in a selection of real world problems, simple random mutation performs better. The paper then finishes with an analysis of the diversity of constants in the population, and the manner in which this changes over time. © Springer-Verlag Berlin Heidelberg 2003.