Benchmarking the novel CMA-ES restart strategy using the search history on the BBOB noiseless testbed
Takahiro Yamaguchi, Youhei Akimoto · Proceedings of the Genetic and Evolutionary Computation Conference Companion · 2017
In this paper we propose a termination mechanism and the initial step-size control mechanism for restart strategies in the CMA-ES. The proposed mechanism utilizes a history of the distribution parameters from past restarts to early terminate an overlapping exploitation of the search domain. The initial step-size is controlled so that the next restart will not overlap with past restarts. The proposed mechanism is combined with a simple restart, IPOP restart and BIPOP restart strategies. The effectiveness and the drawback of the proposed mechanism is demonstrated on the BBOB noiseless testbed.