Scaling Up Reinforcement Learning through Targeted Exploration