GPU-Accelerated Standard and Multi-population Cultural Algorithms

Abstract

In this paper, we present three parallel cultural algorithms using CUDA-enabled GPUs. Firstly, we used the GPU to accelerate an expensive fitness function. Next, the parallel versions of both standard and multi-population CAs were presented. Experiments show that the standard CA with an expensive fitness function was made more than 600 times faster. On lightweight benchmark problems, the speedups were only 3-4 times for the standard CA while the multi-population CA can still achieve 30-50 times speedups.