Ring Optimization with Extinction

Daniel Ashlock, Sierra Gillis, and Gray Fogel
Accepted to the 2015 Congress on Evolutionary Computation

Abstract PDF eprint

Extinction is a natural process that drives biological evolution. In this study the impact of introducing extinction operators into ring optimization was examined. Ring optimizers are spatially structured evolutionary optimizers inspired by the biological phenomenon of a ring species. A small initial population is introduced into a ring-structured space and spreads, using the spatial structure to manage the exploration/exploitation trade-off of the algorithm. Extinction operators eliminate a substantial fraction of the current population, in effect resetting the algorithm to a more exploratory state. Two types of extinction operators are tested and compared. The "deluge operator" removes population members with lower fitness while the "asteroid operator" removes population members in a contiguous block of the ring. Three benchmark functions were used, one a discrete simulation and the other two open-ended continuous real functions. The behavior of the extinction operators are different for each of the benchmark functions. The differences in behavior of the extinction operators are explained in terms of the fitness landscapes of the benchmark functions.