GPGPU-compatible archive based stochastic ranking evolutionary algorithm (G-ASREA) for multi-objective optimization
LogXlabs Research Center, Paris, France
Proceedings of the 11th international conference on Parallel problem solving from nature: Part II, PPSN’10, p.111-120
@article{sharmagpgpu,
title={GPGPU-Compatible Archive Based Stochastic Ranking Evolutionary Algorithm (G-ASREA) for Multi-Objective Optimization},
author={Sharma, D. and Collet, P.},
journal={Parallel Problem Solving from Nature–PPSN XI},
pages={111–120},
publisher={Springer}
}
In this paper, a GPGPU (general purpose graphics processing unit) compatible Archived based Stochastic Ranking Evolutionary Algorithm (G-ASREA) is proposed, that ranks the population with respect to an archive of non-dominated solutions. It reduces the complexity of the deterministic ranking operator from O(mn^2) to O(man)* and further speeds up ranking on GPU. Experiments compare G-ASREA with a CPU version of ASREA and NSGA-II on ZDT test functions for a wide range of population sizes. The results confirm the gain in ranking complexity by showing that on 10K individuals, G-ASREA ranking is ~x5000 faster than NSGA-II and ~x15 faster than ASREA.
February 4, 2011 by hgpu
