Particle Swarm Optimization on a GPU
Mikhail I. Rabinovich, Phillip Kainga, David Johnson, Brandon Shafer, John J. Lee, R.C. Eberhart · 2012
Optimization problems that contain discontinuities, non-linearity, or high dimensionality are difficult to solve and time consuming using conventional computational methods. This paper introduces a tool that solves these kinds of optimization problems using a patent pending Gaming Particle Swarm Optimization (GPSO) algorithm implemented on Graphics Processing Unit (GPU) hardware. Our study applied this utility to a radio frequency resource allocation optimizer. This tool, implemented on an Nvidia GTX 465, resulted in 5X performance gain over a state-of-the-art AMD Phenom 3.4GHz quad-core CPU. This study provides a powerful tool that may be used for solving various multi-disciplinary optimization problems such as training of artificial neural networks, function maximization/minimization, autotuning for universal mobile telecommunication system networks, as well as scheduling.