Adaptive Representations for Reinforcement Learning

The first approach synthesizes temporal difference methods, the traditional approach to reinforcement learning, with evolutionary methods, which can learn representations for a broad class ofoptimization problems.