Ensemble and Auxiliary Tasks for Data-Efficient Deep Reinforcement Learning

Ensemble and auxiliary tasks are both well known to improve the performance\nof machine learning models when data is limited. However, the interaction\nbetween these two methods is not well studied, particularly in the context of\ndeep reinforcement learning. In this paper, we study the effects of ensemble\nand auxiliary tasks when combined with the deep Q-learning algorithm. We\nperform a case study on ATARI games under limited data constraint. Moreover, we\nderive a refined bias-variance-covariance decomposition to analyze the\ndifferent ways of learning ensembles and using auxiliary tasks, and use the\nanalysis to help provide some understanding of the case study. Our code is open\nsource and available at https://github.com/NUS-LID/RENAULT.\n