EX2: Exploration with Exemplar Models for Deep Reinforcement Learning