Asynchronous island parallel GA using multiform subpopulations

Abstract

Island Parallel GA divides a population into subpopulations and assigns them to processing elements on a parallel computer. Then each subpopulation searches the optimal solution independently, and ex-changes individuals periodically. This exchange operation is called mi-gration. In this research, we propose a new algorithm that migrants are exchanged asynchronously among multiform subpopulations which have different search conditions. The effect of our algorithm on combina-tional optimization problems was verified by applying the algorithm to Knapsack Problem and Royal Road Functions using parallel computer CRAY-T3E. We obtained the results that our algorithm maintained the population’s diversity effectively and searches building blocks efficiently.