Topologies, Migration Rates, and Multi-Population Parallel Genetic Algorithms

This paper presents a study of parallel genetic algorithms (GAs) with multiple populations (also called demes or islands). The study makes explicit the relation between the probability of reaching a desired solution with the deme size, the migration rate, and the degree of the connectivity graph. The paper considers arbitrary topologies with a xed number of neighbors per deme. The demes evolve in isolation until each converges to a unique solution. Then, the demes exchange an arbitrary number of individuals and restart their execution. An accurate deme-sizing equation is derived, and it is used to determine the optimal con guration of an arbitrary number of demes that minimizes the execution time of the parallel GA.

• Type: Conference
• Year: 1999
• Where: GECCO
• Authors: Erick Cantú-Paz
• Keywords: Arbitrary Number, Demes, GECCO 1999, Optimization, Parallel Genetic Algorithms