Ensemble and Auxiliary Tasks for Data-Efficient Deep Reinforcement Learning