結果 : parallel reinforcement learning with linear function approximation