A functional specialization hypothesis for designing genetic algorithms

Intensive studies of genetic algorithms (GAs) make the GAs really effective techniques applicable to hard optimization problems. These studies suggest two key points in designing optimizers using the GAs: first, GAs should be designed so as to maintain the diversity of the population well; second, they should be designed so as to inherit good characteristics from parents well. The paper tries to make these observations more concrete guidelines for designing GAs. First, an alternative picture that captures the search process of the GA as evolution of the probability distribution function of the population is proposed. Then, based on this picture, a functional specialization hypothesis that specifies the roles of selection and crossover operators is proposed as guidelines to design GAs. Then, the state-of-the-art selection and crossover operators for continuous search spaces are introduced along the proposed guidelines. Further, crossover operators for discrete search spaces are also discussed from this viewpoint.