Instance-Optimality in Interactive Decision Making: Toward a Non-Asymptotic Theory

Andrew J. Wagenmaker, Dylan J. Foster
Proceedings of Thirty Sixth Conference on Learning Theory, PMLR 195:1322-1472, 2023.

Abstract

We consider the development of adaptive, instance-dependent algorithms for interactive decision making (bandits, reinforcement learning, and beyond) that, rather than only performing well in the worst case, adapt to favorable properties of real-world instances for improved performance. We aim for instance-optimality, a strong notion of adaptivity which asserts that, on any particular problem instance, the algorithm under consideration outperforms all consistent algorithms. Instance-optimality enjoys a rich asymptotic theory originating from the work of \citet{lai1985asymptotically} and \citet{graves1997asymptotically}, but non-asymptotic guarantees have remained elusive outside of certain special cases. Even for problems as simple as tabular reinforcement learning, existing algorithms do not attain instance-optimal performance until the number of rounds of interaction is doubly exponential in the number of states.In this paper, we take the first step toward developing a non-asymptotic theory of instance-optimal decision making with general function approximation. We introduce a new complexity measure, the Allocation-Estimation Coefficient (AEC), and provide a new algorithm, AE2, which attains non-asymptotic instance-optimal performance at a rate controlled by the AEC. Our results recover the best known guarantees for well-studied problems such as finite-armed and linear bandits and, when specialized to tabular reinforcement learning, attain the first instance-optimal regret bounds with polynomial dependence on all problem parameters, improving over prior work exponentially. We complement these results with lower bounds that show that i) existing notions of statistical complexity are insufficient to derive non-asymptotic guarantees, and ii) under certain technical conditions, boundedness of the Allocation-Estimation Coefficient is necessary to learn an instance-optimal allocation of decisions in finite time.

Cite this Paper

BibTeX
@InProceedings{pmlr-v195-wagenmaker23a, title = {Instance-Optimality in Interactive Decision Making: Toward a Non-Asymptotic Theory}, author = {Wagenmaker, Andrew J. and Foster, Dylan J.}, booktitle = {Proceedings of Thirty Sixth Conference on Learning Theory}, pages = {1322--1472}, year = {2023}, editor = {Neu, Gergely and Rosasco, Lorenzo}, volume = {195}, series = {Proceedings of Machine Learning Research}, month = {12--15 Jul}, publisher = {PMLR}, pdf = {https://proceedings.mlr.press/v195/wagenmaker23a/wagenmaker23a.pdf}, url = {https://proceedings.mlr.press/v195/wagenmaker23a.html}, abstract = {We consider the development of adaptive, instance-dependent algorithms for interactive decision making (bandits, reinforcement learning, and beyond) that, rather than only performing well in the worst case, adapt to favorable properties of real-world instances for improved performance. We aim for instance-optimality, a strong notion of adaptivity which asserts that, on any particular problem instance, the algorithm under consideration outperforms all consistent algorithms. Instance-optimality enjoys a rich asymptotic theory originating from the work of \citet{lai1985asymptotically} and \citet{graves1997asymptotically}, but non-asymptotic guarantees have remained elusive outside of certain special cases. Even for problems as simple as tabular reinforcement learning, existing algorithms do not attain instance-optimal performance until the number of rounds of interaction is doubly exponential in the number of states.In this paper, we take the first step toward developing a non-asymptotic theory of instance-optimal decision making with general function approximation. We introduce a new complexity measure, the Allocation-Estimation Coefficient (AEC), and provide a new algorithm, AE2, which attains non-asymptotic instance-optimal performance at a rate controlled by the AEC. Our results recover the best known guarantees for well-studied problems such as finite-armed and linear bandits and, when specialized to tabular reinforcement learning, attain the first instance-optimal regret bounds with polynomial dependence on all problem parameters, improving over prior work exponentially. We complement these results with lower bounds that show that i) existing notions of statistical complexity are insufficient to derive non-asymptotic guarantees, and ii) under certain technical conditions, boundedness of the Allocation-Estimation Coefficient is necessary to learn an instance-optimal allocation of decisions in finite time.} }
Endnote
%0 Conference Paper %T Instance-Optimality in Interactive Decision Making: Toward a Non-Asymptotic Theory %A Andrew J. Wagenmaker %A Dylan J. Foster %B Proceedings of Thirty Sixth Conference on Learning Theory %C Proceedings of Machine Learning Research %D 2023 %E Gergely Neu %E Lorenzo Rosasco %F pmlr-v195-wagenmaker23a %I PMLR %P 1322--1472 %U https://proceedings.mlr.press/v195/wagenmaker23a.html %V 195 %X We consider the development of adaptive, instance-dependent algorithms for interactive decision making (bandits, reinforcement learning, and beyond) that, rather than only performing well in the worst case, adapt to favorable properties of real-world instances for improved performance. We aim for instance-optimality, a strong notion of adaptivity which asserts that, on any particular problem instance, the algorithm under consideration outperforms all consistent algorithms. Instance-optimality enjoys a rich asymptotic theory originating from the work of \citet{lai1985asymptotically} and \citet{graves1997asymptotically}, but non-asymptotic guarantees have remained elusive outside of certain special cases. Even for problems as simple as tabular reinforcement learning, existing algorithms do not attain instance-optimal performance until the number of rounds of interaction is doubly exponential in the number of states.In this paper, we take the first step toward developing a non-asymptotic theory of instance-optimal decision making with general function approximation. We introduce a new complexity measure, the Allocation-Estimation Coefficient (AEC), and provide a new algorithm, AE2, which attains non-asymptotic instance-optimal performance at a rate controlled by the AEC. Our results recover the best known guarantees for well-studied problems such as finite-armed and linear bandits and, when specialized to tabular reinforcement learning, attain the first instance-optimal regret bounds with polynomial dependence on all problem parameters, improving over prior work exponentially. We complement these results with lower bounds that show that i) existing notions of statistical complexity are insufficient to derive non-asymptotic guarantees, and ii) under certain technical conditions, boundedness of the Allocation-Estimation Coefficient is necessary to learn an instance-optimal allocation of decisions in finite time.
APA
Wagenmaker, A.J. & Foster, D.J.. (2023). Instance-Optimality in Interactive Decision Making: Toward a Non-Asymptotic Theory. Proceedings of Thirty Sixth Conference on Learning Theory, in Proceedings of Machine Learning Research 195:1322-1472 Available from https://proceedings.mlr.press/v195/wagenmaker23a.html.

Related Material