Empirical runtime distributions on multimodal structured functions (f15: Rastrigin, f16: Weierstrass, f17: Schaffers F7, f18: Moderately ill-conditioned Schaffers F7, f19: Composite Griewank-Rosenbrock) with target values in {100, …, 1e-8} in dimension 20.
The following algorithms perform particularly well up to their individual maximum number of function evaluations: IPOP-SEP-CMA-ES (CMA-ES restarted with Increasing POPulation size, first run with SEParable CMA) and BIPOP-CMA-ES (CMA-ES restarted with budgets for small and large population size).