Distributed NSGA-II with migration using compensation on many-core processors for improving performance and accuracy
Yuji Sato, Mikiko Sato, Minami Miyakawa · Proceedings of the Genetic and Evolutionary Computation Conference Companion · 2017
A recent trend in multiobjective evolutionary algorithms is to increase the population size to approximate the nondominated solution set with high accuracy. And the execution time becomes a problem in engineering applications. In this paper, we propose distributed, high-speed NSGA-II using a many-core environment to obtain a Pareto-optimal solution set excelling in convergence and diversity. This method improves performance while maintaining the accuracy of the Pareto-optimal solution set by repeating NSGA-II distributed processing in a many-core environment inspired by the divide-and-conquer method together with migration processing for compensation of the nondominated solution set obtained by distributed processing. On comparing with NSGA-II executing on a single CPU and parallel, high-speed NSGA-II using a standard island model, it was found that the proposed method greatly shortened the execution time for obtaining a Pareto-optimal solution set with equivalent hypervolume while increasing the accuracy of solution searching.