Adaptive Sample-Efficient Blackbox Optimization via ES-active Subspaces

We present a new algorithm ASEBO for conducting optimization of high-dimensional blackbox functions. ASEBO adapts to the geometry of the function and learns optimal sets of sensing directions, which are used to probe it, on-the-fly. It addresses the exploration-exploitation trade-off of blackbox optimization, where each single function query is expensive, by continuously learning the bias of the lower-dimensional model used to approximate gradients of smoothings of the function with compressed sensing and contextual bandits methods. To obtain this model, it uses techniques from the emerging theory of active subspaces in the novel ES blackbox optimization context. As a result, ASEBO learns the dynamically changing intrinsic dimensionality of the gradient space and adapts to the hardness of different stages of the optimization without external supervision. Consequently, it leads to more sample-efficient blackbox optimization than state-of-the-art algorithms. We provide rigorous theoretical justification of the effectiveness of our method. We also empirically evaluate it on the set of reinforcement learning policy optimization tasks as well as functions from the recently open-sourced Nevergrad library, demonstrating that it consistently learns optimal inputs with fewer queries to a blackbox function than other methods.