ASYNCHRONOUS PARALLEL ALGORITHMS FOR STRATEGIC HYBRID SEARCHING BASED ON A MIXTURE GAUSSIAN MODEL

摘要

Technically exploring a solution space in an effective way helps not only to find good quality solutions, but also to reduce computation time. This paper proposes an optimization technique that utilizes hybridization, strategic search, parallelization, and asynchronous cooperation. A master-slave topology has been formulated in which the master strategically sorts out portions of the search space in four phases with the help of a clustering algorithm and assumes the role of an estimation of distribution algorithm to model the solution distribution within the space using a Gaussian mixture model without variable dependency. The new algorithm models a solution distribution by considering not only the mean vector of clustered solutions obtained from previous searches, as per the continuous univariate marginal distribution algorithm, but also by including information about the quality of solutions. With sorted probability distributions assigned by the master, slaves use genetic algorithms to extensively explore the solution space. The effect of our proposal has been experimentally analyzed in continuous domains, and the resultant algorithm shows significant improvements both in finding relatively good solutions and in reducing computation time.