Using holey fitness landscapes to counteract premature convergence in evolutionary algorithms

Abstract

Premature convergence is a persisting problem in evolutionary optimisation, in particular - genetic algorithms. While a number of methods exist to approach this issue, they usually require problem specific calibration or only partially resolve the issue, at best by delaying the premature convergence of an evolving population. Analytical models in biology show that resiliently diverse populations evolve on high-dimensional fitness landscapes with rather than rugged topographies, but the implications for artificial evolutionary systems remain largely unexplored. Here I show how holey fitness landscapes (HFLs) can be incorporated in an evolutionary algorithm and use this approach to investigate the ability of HFLs to maintain genetic diversity in an evolving population. The results indicate that an underlying HFL can counteract premature genetic convergence and sustain diversity. They also suggest that HFL may provide a flexible mechanism for dynamic creation and maintenance of subpopulations that concentrate their evolutionary search in different regions of the solution space. Finally, I discuss on-going work on using the HFL model in optimisation problems.