State Abstractions for Lifelong Reinforcement Learning

David Abel, Dilip Arumugam, Lucas Lehnert, Michael Littman
Proceedings of the 35th International Conference on Machine Learning, PMLR 80:10-19, 2018.

Abstract

In lifelong reinforcement learning, agents must effectively transfer knowledge across tasks while simultaneously addressing exploration, credit assignment, and generalization. State abstraction can help overcome these hurdles by compressing the representation used by an agent, thereby reducing the computational and statistical burdens of learning. To this end, we here develop theory to compute and use state abstractions in lifelong reinforcement learning. We introduce two new classes of abstractions: (1) transitive state abstractions, whose optimal form can be computed efficiently, and (2) PAC state abstractions, which are guaranteed to hold with respect to a distribution of tasks. We show that the joint family of transitive PAC abstractions can be acquired efficiently, preserve near optimal-behavior, and experimentally reduce sample complexity in simple domains, thereby yielding a family of desirable abstractions for use in lifelong reinforcement learning. Along with these positive results, we show that there are pathological cases where state abstractions can negatively impact performance.

Cite this Paper


BibTeX
@InProceedings{pmlr-v80-abel18a, title = {State Abstractions for Lifelong Reinforcement Learning}, author = {Abel, David and Arumugam, Dilip and Lehnert, Lucas and Littman, Michael}, booktitle = {Proceedings of the 35th International Conference on Machine Learning}, pages = {10--19}, year = {2018}, editor = {Dy, Jennifer and Krause, Andreas}, volume = {80}, series = {Proceedings of Machine Learning Research}, month = {10--15 Jul}, publisher = {PMLR}, pdf = {http://proceedings.mlr.press/v80/abel18a/abel18a.pdf}, url = {https://proceedings.mlr.press/v80/abel18a.html}, abstract = {In lifelong reinforcement learning, agents must effectively transfer knowledge across tasks while simultaneously addressing exploration, credit assignment, and generalization. State abstraction can help overcome these hurdles by compressing the representation used by an agent, thereby reducing the computational and statistical burdens of learning. To this end, we here develop theory to compute and use state abstractions in lifelong reinforcement learning. We introduce two new classes of abstractions: (1) transitive state abstractions, whose optimal form can be computed efficiently, and (2) PAC state abstractions, which are guaranteed to hold with respect to a distribution of tasks. We show that the joint family of transitive PAC abstractions can be acquired efficiently, preserve near optimal-behavior, and experimentally reduce sample complexity in simple domains, thereby yielding a family of desirable abstractions for use in lifelong reinforcement learning. Along with these positive results, we show that there are pathological cases where state abstractions can negatively impact performance.} }
Endnote
%0 Conference Paper %T State Abstractions for Lifelong Reinforcement Learning %A David Abel %A Dilip Arumugam %A Lucas Lehnert %A Michael Littman %B Proceedings of the 35th International Conference on Machine Learning %C Proceedings of Machine Learning Research %D 2018 %E Jennifer Dy %E Andreas Krause %F pmlr-v80-abel18a %I PMLR %P 10--19 %U https://proceedings.mlr.press/v80/abel18a.html %V 80 %X In lifelong reinforcement learning, agents must effectively transfer knowledge across tasks while simultaneously addressing exploration, credit assignment, and generalization. State abstraction can help overcome these hurdles by compressing the representation used by an agent, thereby reducing the computational and statistical burdens of learning. To this end, we here develop theory to compute and use state abstractions in lifelong reinforcement learning. We introduce two new classes of abstractions: (1) transitive state abstractions, whose optimal form can be computed efficiently, and (2) PAC state abstractions, which are guaranteed to hold with respect to a distribution of tasks. We show that the joint family of transitive PAC abstractions can be acquired efficiently, preserve near optimal-behavior, and experimentally reduce sample complexity in simple domains, thereby yielding a family of desirable abstractions for use in lifelong reinforcement learning. Along with these positive results, we show that there are pathological cases where state abstractions can negatively impact performance.
APA
Abel, D., Arumugam, D., Lehnert, L. & Littman, M.. (2018). State Abstractions for Lifelong Reinforcement Learning. Proceedings of the 35th International Conference on Machine Learning, in Proceedings of Machine Learning Research 80:10-19 Available from https://proceedings.mlr.press/v80/abel18a.html.

Related Material