Sample-Efficient Reinforcement Learning