Evolutionary multi-objective optimization algorithms aim at finding an approximation of the set of Pareto optimal solutions. In the case of complex problems with many conflicting objectives, a lot of non-dominated solutions may be obtained to represent the Pareto front. It is desirable to produce the Pareto front in a reasonable amount of time. Parallel implementations of genetic algorithms are developed to speed-up the solution process. However, most of studies are focused on parallelization of a particular algorithm. In our research, we aim to speed-up the most time consuming procedure, non-dominated sorting, which is used in several state-of-the-art multi-objective genetic algorithms.
Two versions of non-dominated sorting procedure can be downloaded: (1) an improved sequential in MATLAB; (2) a hybrid code (based on Pthreads and CUDA).
Authors: G. Ortega1, E. Filatovas2, E. M. Garzón1 and L. G. Casado1
1Dpt Computer Architecture. Almeria University (ceiA3), Ctra Sacramento s/n Almeria, 04120 Spain.
2Institute of Informatics and Mathematics, Vilnius University, Akademijos str. 4, LT: 08663 Vilnius, Lithuania