Benchmarking some variants of the CMAES-APOP using keeping search points and mirrored sampling combined with active CMA on the BBOB noiseless testbed
Duc Manh Nguyen · Proceedings of the Genetic and Evolutionary Computation Conference Companion · 2022
In this paper, we evaluate some variants of the CMAES-APOP using two additional methods. The first method is the combination of mirrored sampling with active covariance matrix adaptation. The second one is to keep some search points of the previous iteration for the evolution of the current iteration when the population size is large. Moreover, we propose a strategy to reduce the step-size when the population size attains its upper bound for several consecutive iterations. We will show that the techniques can enhance the performance on some multi-modal functions, especially on the weakly-structured multi-modal ones, and the overall performance in all dimensions.