Variance Double-Down: The Small Batch Size Anomaly in Multistep Deep Reinforcement Learning

Johan Samir Obando Ceron, Marc G Bellemare, Pablo Samuel Castro

Keywords: Reinforcement Learning, Deep Reinforcement Learning, Value based, Batch Size, Multi step learning

TL;DR: We perform an exhaustive investigation into the interplay of batch size and update horizon and uncover a surprising phenomenon: when increasing the update horizon, it is more beneficial to decrease the batch size

Abstract: State of the art results in reinforcement learning suggest that multi-step learning is necessary. However, the increased variance that comes with it makes it difficult to increase the update horizon beyond relatively small numbers. In this paper, we report the counterintuitive finding that decreasing the batch size substantially improves performance across a large swath of deep RL agents. It is well-known that gradient variance decreases with increasing batch sizes, so obtaining improved performance by increasing variance on two fronts is a rather surprising finding. We conduct a broad set of experiments to better understand this variance double-down phenomenon.

Anonymous Url: I certify that there is no URL (e.g., github page) that could be used to find authors’ identity.

No Acknowledgement Section: I certify that there is no acknowledgement section in this submission for double blind review.

Code Of Ethics: I acknowledge that I and all co-authors of this work have read and commit to adhering to the ICLR Code of Ethics

Submission Guidelines: Yes

Please Choose The Closest Area That Your Submission Falls Into: Reinforcement Learning (eg, decision and control, planning, hierarchical RL, robotics)