A simple bucketing based approach to diversity maintenance

We present an approach to diversity maintenance based on separating the population into buckets based on similarity and biasing selection to keep individuals from all buckets in the population. We look at two approaches to bucketing. The first uses a locally sensitive bucketing function on individuals. The second uses the K-Means clustering algorithms to divide the population. We focus our research on a family of deceptive problem domains which we dub Tricky Keys and analyze how the using bucketing methods changes evolutionary search results for problem instances of varying diffculty. Our results show that both bucketing by function and bucketing by clustering methods show an increase in probability of finding a good solution and in number of good solutions found.