Offline Meta Reinforcement Learning with In-Distribution Online Adaptation

Jianhao Wang, Jin Zhang, Haozhe Jiang, Junyu Zhang, Liwei Wang, Chongjie Zhang
Proceedings of the 40th International Conference on Machine Learning, PMLR 202:36626-36669, 2023.

Abstract

Recent offline meta-reinforcement learning (meta-RL) methods typically utilize task-dependent behavior policies (e.g., training RL agents on each individual task) to collect a multi-task dataset. However, these methods always require extra information for fast adaptation, such as offline context for testing tasks. To address this problem, we first formally characterize a unique challenge in offline meta-RL: transition-reward distribution shift between offline datasets and online adaptation. Our theory finds that out-of-distribution adaptation episodes may lead to unreliable policy evaluation and that online adaptation with in-distribution episodes can ensure adaptation performance guarantee. Based on these theoretical insights, we propose a novel adaptation framework, called In-Distribution online Adaptation with uncertainty Quantification (IDAQ), which generates in-distribution context using a given uncertainty quantification and performs effective task belief inference to address new tasks. We find a return-based uncertainty quantification for IDAQ that performs effectively. Experiments show that IDAQ achieves state-of-the-art performance on the Meta-World ML1 benchmark compared to baselines with/without offline adaptation.

Cite this Paper

BibTeX
@InProceedings{pmlr-v202-wang23au, title = {Offline Meta Reinforcement Learning with In-Distribution Online Adaptation}, author = {Wang, Jianhao and Zhang, Jin and Jiang, Haozhe and Zhang, Junyu and Wang, Liwei and Zhang, Chongjie}, booktitle = {Proceedings of the 40th International Conference on Machine Learning}, pages = {36626--36669}, 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/wang23au/wang23au.pdf}, url = {https://proceedings.mlr.press/v202/wang23au.html}, abstract = {Recent offline meta-reinforcement learning (meta-RL) methods typically utilize task-dependent behavior policies (e.g., training RL agents on each individual task) to collect a multi-task dataset. However, these methods always require extra information for fast adaptation, such as offline context for testing tasks. To address this problem, we first formally characterize a unique challenge in offline meta-RL: transition-reward distribution shift between offline datasets and online adaptation. Our theory finds that out-of-distribution adaptation episodes may lead to unreliable policy evaluation and that online adaptation with in-distribution episodes can ensure adaptation performance guarantee. Based on these theoretical insights, we propose a novel adaptation framework, called In-Distribution online Adaptation with uncertainty Quantification (IDAQ), which generates in-distribution context using a given uncertainty quantification and performs effective task belief inference to address new tasks. We find a return-based uncertainty quantification for IDAQ that performs effectively. Experiments show that IDAQ achieves state-of-the-art performance on the Meta-World ML1 benchmark compared to baselines with/without offline adaptation.} }
Endnote
%0 Conference Paper %T Offline Meta Reinforcement Learning with In-Distribution Online Adaptation %A Jianhao Wang %A Jin Zhang %A Haozhe Jiang %A Junyu Zhang %A Liwei Wang %A Chongjie Zhang %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-wang23au %I PMLR %P 36626--36669 %U https://proceedings.mlr.press/v202/wang23au.html %V 202 %X Recent offline meta-reinforcement learning (meta-RL) methods typically utilize task-dependent behavior policies (e.g., training RL agents on each individual task) to collect a multi-task dataset. However, these methods always require extra information for fast adaptation, such as offline context for testing tasks. To address this problem, we first formally characterize a unique challenge in offline meta-RL: transition-reward distribution shift between offline datasets and online adaptation. Our theory finds that out-of-distribution adaptation episodes may lead to unreliable policy evaluation and that online adaptation with in-distribution episodes can ensure adaptation performance guarantee. Based on these theoretical insights, we propose a novel adaptation framework, called In-Distribution online Adaptation with uncertainty Quantification (IDAQ), which generates in-distribution context using a given uncertainty quantification and performs effective task belief inference to address new tasks. We find a return-based uncertainty quantification for IDAQ that performs effectively. Experiments show that IDAQ achieves state-of-the-art performance on the Meta-World ML1 benchmark compared to baselines with/without offline adaptation.
APA
Wang, J., Zhang, J., Jiang, H., Zhang, J., Wang, L. & Zhang, C.. (2023). Offline Meta Reinforcement Learning with In-Distribution Online Adaptation. Proceedings of the 40th International Conference on Machine Learning, in Proceedings of Machine Learning Research 202:36626-36669 Available from https://proceedings.mlr.press/v202/wang23au.html.

Related Material