Efficient Reinforcement Learning in Multiple-Agent Systems and its Application in Cognitive Radio Networks