Parallel genetic algorithms with schema migration

Genetic algorithms (GAs) are efficient non-gradient stochastic search methods. Parallel GAs are proposed to overcome the deficiencies of sequential GAs, such as low speed and aptness to locally converge. However the tremendous communication cost incurred offsets the advantages of parallel GAs. Hence reducing communication cost is the key issue of this problem. Instead of reducing the communication cost simply by compressing the size of the messages, we tackle the problem by improving the effectiveness of the schema to be disseminated. We propose a new schema migration scheme (SMS). This SMS consists of a schema extracting mechanism and a schema disseminating mechanism. This SMS is valid and requires less communication cost.