Illinois Genetic Algorithms Laboratory

Cantú-Paz, E., & Goldberg, D. E (1996)
TR No.: 96007 | Download PDF | Download PS

Abstract:

This paper presents models to predict the quality of convergence of idealized bounding cases of parallel genetic algorithms (GAs). The first bounding case is a parallel GA with completely isolated subpopulations (demes). We show how the probability that the parallel GA finds a solution of the minimum desired quality increases as more demes are used. Our second bounding case is when each deme communicates with every other with a maximal migration rate. We derive a model to predict the probability that a building block converges to the correct value based on a previous model for simple GAs. For each of the bounding cases, we derive equations that determine the deme size that is required when the quality of the solution and the number of demes are fixed.