Reinforcement Learning Can Be More Efficient with Multiple Rewards

Christoph Dann, Yishay Mansour, Mehryar Mohri
Proceedings of the 40th International Conference on Machine Learning, PMLR 202:6948-6967, 2023.

Abstract

Reward design is one of the most critical and challenging aspects when formulating a task as a reinforcement learning (RL) problem. In practice, it often takes several attempts of reward specification and learning with it in order to find one that leads to sample-efficient learning of the desired behavior. Instead, in this work, we study whether directly incorporating multiple alternate reward formulations of the same task in a single agent can lead to faster learning. We analyze multi-reward extensions of action-elimination algorithms and prove more favorable instance-dependent regret bounds compared to their single-reward counterparts, both in multi-armed bandits and in tabular Markov decision processes. Our bounds scale for each state-action pair with the inverse of the largest gap among all reward functions. This suggests that learning with multiple rewards can indeed be more sample-efficient, as long as the rewards agree on an optimal policy. We further prove that when rewards do not agree, multi-reward action elimination in multi-armed bandits still learns a policy that is good across all reward functions.

Cite this Paper

BibTeX
@InProceedings{pmlr-v202-dann23a, title = {Reinforcement Learning Can Be More Efficient with Multiple Rewards}, author = {Dann, Christoph and Mansour, Yishay and Mohri, Mehryar}, booktitle = {Proceedings of the 40th International Conference on Machine Learning}, pages = {6948--6967}, year = {2023}, editor = {Krause, Andreas and Brunskill, Emma and Cho, Kyunghyun and Engelhardt, Barbara and Sabato, Sivan and Scarlett, Jonathan}, volume = {202}, series = {Proceedings of Machine Learning Research}, month = {23--29 Jul}, publisher = {PMLR}, pdf = {https://proceedings.mlr.press/v202/dann23a/dann23a.pdf}, url = {https://proceedings.mlr.press/v202/dann23a.html}, abstract = {Reward design is one of the most critical and challenging aspects when formulating a task as a reinforcement learning (RL) problem. In practice, it often takes several attempts of reward specification and learning with it in order to find one that leads to sample-efficient learning of the desired behavior. Instead, in this work, we study whether directly incorporating multiple alternate reward formulations of the same task in a single agent can lead to faster learning. We analyze multi-reward extensions of action-elimination algorithms and prove more favorable instance-dependent regret bounds compared to their single-reward counterparts, both in multi-armed bandits and in tabular Markov decision processes. Our bounds scale for each state-action pair with the inverse of the largest gap among all reward functions. This suggests that learning with multiple rewards can indeed be more sample-efficient, as long as the rewards agree on an optimal policy. We further prove that when rewards do not agree, multi-reward action elimination in multi-armed bandits still learns a policy that is good across all reward functions.} }
Endnote
%0 Conference Paper %T Reinforcement Learning Can Be More Efficient with Multiple Rewards %A Christoph Dann %A Yishay Mansour %A Mehryar Mohri %B Proceedings of the 40th International Conference on Machine Learning %C Proceedings of Machine Learning Research %D 2023 %E Andreas Krause %E Emma Brunskill %E Kyunghyun Cho %E Barbara Engelhardt %E Sivan Sabato %E Jonathan Scarlett %F pmlr-v202-dann23a %I PMLR %P 6948--6967 %U https://proceedings.mlr.press/v202/dann23a.html %V 202 %X Reward design is one of the most critical and challenging aspects when formulating a task as a reinforcement learning (RL) problem. In practice, it often takes several attempts of reward specification and learning with it in order to find one that leads to sample-efficient learning of the desired behavior. Instead, in this work, we study whether directly incorporating multiple alternate reward formulations of the same task in a single agent can lead to faster learning. We analyze multi-reward extensions of action-elimination algorithms and prove more favorable instance-dependent regret bounds compared to their single-reward counterparts, both in multi-armed bandits and in tabular Markov decision processes. Our bounds scale for each state-action pair with the inverse of the largest gap among all reward functions. This suggests that learning with multiple rewards can indeed be more sample-efficient, as long as the rewards agree on an optimal policy. We further prove that when rewards do not agree, multi-reward action elimination in multi-armed bandits still learns a policy that is good across all reward functions.
APA
Dann, C., Mansour, Y. & Mohri, M.. (2023). Reinforcement Learning Can Be More Efficient with Multiple Rewards. Proceedings of the 40th International Conference on Machine Learning, in Proceedings of Machine Learning Research 202:6948-6967 Available from https://proceedings.mlr.press/v202/dann23a.html.

Related Material