Learning state correspondence of reinforcement learning tasks for knowledge transfer