發送短信: Efficient PAC reinforcement learning in regular decision processes