Evolving Fitness Functions for Mating Selection

The tailoring of an evolutionary algorithm to a specific problem is typically a time-consuming and complex process. Over the years, several approaches have been proposed for the automatic adaptation of parameters and components of evolutionary algorithms. We focus on the evolution of mating selection fitness functions and use as case study the Circle Packing in Squares problem. Each individual encodes a potential solution for the circle packing problem and a fitness function, which is used to assess the suitability of its potential mating partners. The experimental results show that by evolving mating selection functions it is possible to surpass the results attained with hardcoded fitness functions. Moreover, they also indicate that genetic programming was able to discover mating selection functions that: use the information regarding potential mates in novel and unforeseen ways; outperform the class of mating functions considered by the authors.