Improving reinforcement learning with ensembles of different learners

Abstract

Different reinforcement learning (RL) methods exist to address the problem of combining multiple different learners to generate a superior learner. These existing methods usually assume that each learner uses the same algorithm and/or state representation. We propose an ensemble learner that combines a set of base learners and leverages the strengths of the different base learners online. We demonstrate the proposed ensemble learner’s ability to combine the strengths of multiple base learners and adapt to changes in base learner performance on various domains, including the Atari Breakout domain.