A Study on a Parallelization of Parameter-Less Population Pyramid

Taku Hasegawa, Tomoharu Nakashima

Abstract: In this work, we explore efficient parallelization methods for the Parameter-less Population Pyramid (P3) and show their computational effectiveness. First, in this paper, we present one possible implementation of parallel P3. Next, we analyze effects of sharing solutions between sub-populations of parallel P3 on some benchmarks. By sharing solutions, the parallel P3 finds optimal solutions faster than the simple parallelization. On the deceptive trap and discretized Rastrigin problems, the more solutions parallel P3 shares among sub-populations, the faster it finds optimal solutions. On the other hand, on the Nk problem, too much sharing prevents parallel P3 from performing good search. Finally, we propose guidelines for future research on parallel P3.