Tightening Exploration in Upper Confidence Reinforcement Learning