Adaptive stochastic optimization using multiprocessors (Q1907087)

An adaptive random search algorithm for high-order systems, which periodically optimizes the variance of the step-size distribution, is presented by making use of parallel computation, where processors are collaborated in searching for the optimum solution, communicating periodically with the supervising processor to report on the current best value. Analytical expressions that establish the interaction effect between the magnitude of the random step size, the deviation of the current value from its optimum value and the probability of improving the value of the criterion function, are presented for the case with uniform and Gauss distributions.