Boosting Reinforcement Learning with Extremum Experiences

Ezgi Korkmaz

Primary Area: reinforcement learning

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.

Keywords: Experience Collection, Canonicalization, Environment Interaction, Consistent Improvement over the Baseline

Submission Guidelines: I certify that this submission complies with the submission instructions as described on https://iclr.cc/Conferences/2024/AuthorGuide.

Abstract: Reinforcement learning research has achieved high acceleration in its progress starting from the initial installation of deep neural networks as function approximators to learn policies that make sequential decisions in high-dimensional state representation MDPs. While several consecutive barriers have been broken in deep reinforcement learning research (i.e. learning from high-dimensional states, learning purely via self-play), several others still stand. On this line, in our paper we focus on experience collection in high-dimensional complex MDPs and we propose a unique technique based on experiences obtained through extremum actions. Our method provides theoretical basis for efficient experience collection, and further comes with zero additional computational cost while leading to significant sample efficiency gains in deep reinforcement learning training. We conduct extensive experiments in the Arcade Learning Environment with high-dimensional state representation MDPs. We demonstrate that our technique improves the human normalized median scores of Arcade Learning Environment by 248% in the low-data regime.

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.

Submission Number: 7623