This sort of approach is also used in reinforcement learning https://arxiv.org/abs/1702.01182