Abstract: Many science and engineering applications feature non-convex optimization problems where the performance is not explicitly modeled by a cost or reward function, i.e. it is a black box.