circuit basis of reinforcement learning