Evolving dynamic fitness measures for genetic programming

Abstract

This research builds on the hypothesis that the use of different fitness measures on the different generations of genetic programming (GP) is more effective than the convention of applying the same fitness measure individually throughout GP. Whereas the previous study used a genetic algorithm (GA) to induce the sequence in which fitness measures should be applied over the GP generations, this research uses a meta- (or high-level) GP to evolve a combination of the fitness measures for the low-level GP. The study finds that the meta-GP is the preferred approach to generating dynamic fitness measures. GP systems applying the generated dynamic fitness measures consistently outperform the previous approach, as well as standard GP on benchmark and real world problems. Furthermore, the generated dynamic fitness measures are shown to be reusable, whereby they can be used to solve unseen problems to optimality.