MPGA - a parallel, multi-population genetic algorithm

Author: Erik Rauch

MPGA is a parallel GA for function optimization that runs on any parallel setup supported by Linda. It maintains separate populations of solutions - one on each processor - with occasional migration between them. It was found that this is helpful in maintaining diversity among the solutions, and thus avoiding getting stuck in a local minimum.

MPGA was originally written in 1993 as part of a computational biology project and is being used to optimize 60 floating-point variables in simulating a connectionist model of the development of the fruit-fly embryo. On this problem, it was significantly faster than a simulated annealer.

For more information, email the author.