Niches in NK-Landscapes

Abstract Introduced by Kauffman in 1989, NK-landscapes have been the focus of numerous theoretical and empirical studies in the field of evolutionary computation. Despite all that has been learned from these studies, there are still many open questions concerning NK-landscapes. Most of these studies have been performed using very small problems and have neglected to benchmark the performances of genetic algorithms (GA) with those of hill-climbers, leaving us to wonder if a GA would be the best tool for solving any size NK-landscape problem. Heckendorn, Rana, and Whitley [7] performed initial investigations addressing these questions for NK-landscapes where N  = 100, concluding that an enhanced random bit-climber was best for solving NK-landscapes. Replicating and extending their work, we conclude that a niche exists for GAs like CHC in the NK-landscape functions and describe the bounds of this niche. We also offer some explanations for these bounds and speculate about how the bounds might change as the NK-landscape functions become larger.