Policy Learning with Abstention

Ayush Sawarni, Jikai Jin, Justin Whitehouse, Vasilis Syrgkanis
Proceedings of The 29th International Conference on Artificial Intelligence and Statistics, PMLR 300:4465-4473, 2026.

Abstract

Policy learning algorithms are regularly leveraged in domains such as personalized medicine and advertising to develop individualized treatment regimes. However, a critical deficit of existing algorithms is that they force a decision even when predictions are uncertain, a risky approach in high-stakes settings. The ability to abstain, that is, to defer to a safe default or an expert, is crucial but largely unexplored in this context. To remedy this, we introduce a framework for policy learning with abstention, in which policies that choose not to assign a treatment to some customers/patients receive a small, additive reward on top of the value of a random guess. We propose a two-stage learner that first identifies a set of near-optimal policies and then constructs an abstention class based on disagreements between the policies. We establish fast $O(1/n)$-type regret guarantees for the abstaining policy when propensities are known, and show how to extend these guarantees to the unknown-propensity case via a doubly robust (DR) objective. Furthermore, we demonstrate that our abstention framework is a versatile tool with direct applications to several other core problems in policy learning. We use our algorithm as a black box to obtain improved guarantees under margin conditions without the common realizability assumption. We also show that abstention provides a natural connection to both distributionally robust policy learning, where it acts as a hedge against small data shifts, and safe policy improvement, where the goal is to improve upon a baseline policy with high probability.

Cite this Paper

BibTeX
@InProceedings{pmlr-v300-sawarni26a, title = { Policy Learning with Abstention }, author = {Sawarni, Ayush and Jin, Jikai and Whitehouse, Justin and Syrgkanis, Vasilis}, booktitle = {Proceedings of The 29th International Conference on Artificial Intelligence and Statistics}, pages = {4465--4473}, year = {2026}, editor = {Khan, Emtiyaz and Li, Yingzhen and Solin, Arno and Ramdas, Aaditya}, volume = {300}, series = {Proceedings of Machine Learning Research}, month = {02--05 May}, publisher = {PMLR}, pdf = {https://raw.githubusercontent.com/mlresearch/v300/main/assets/sawarni26a/sawarni26a.pdf}, url = {https://proceedings.mlr.press/v300/sawarni26a.html}, abstract = { Policy learning algorithms are regularly leveraged in domains such as personalized medicine and advertising to develop individualized treatment regimes. However, a critical deficit of existing algorithms is that they force a decision even when predictions are uncertain, a risky approach in high-stakes settings. The ability to abstain, that is, to defer to a safe default or an expert, is crucial but largely unexplored in this context. To remedy this, we introduce a framework for policy learning with abstention, in which policies that choose not to assign a treatment to some customers/patients receive a small, additive reward on top of the value of a random guess. We propose a two-stage learner that first identifies a set of near-optimal policies and then constructs an abstention class based on disagreements between the policies. We establish fast $O(1/n)$-type regret guarantees for the abstaining policy when propensities are known, and show how to extend these guarantees to the unknown-propensity case via a doubly robust (DR) objective. Furthermore, we demonstrate that our abstention framework is a versatile tool with direct applications to several other core problems in policy learning. We use our algorithm as a black box to obtain improved guarantees under margin conditions without the common realizability assumption. We also show that abstention provides a natural connection to both distributionally robust policy learning, where it acts as a hedge against small data shifts, and safe policy improvement, where the goal is to improve upon a baseline policy with high probability. } }
Endnote
%0 Conference Paper %T Policy Learning with Abstention %A Ayush Sawarni %A Jikai Jin %A Justin Whitehouse %A Vasilis Syrgkanis %B Proceedings of The 29th International Conference on Artificial Intelligence and Statistics %C Proceedings of Machine Learning Research %D 2026 %E Emtiyaz Khan %E Yingzhen Li %E Arno Solin %E Aaditya Ramdas %F pmlr-v300-sawarni26a %I PMLR %P 4465--4473 %U https://proceedings.mlr.press/v300/sawarni26a.html %V 300 %X Policy learning algorithms are regularly leveraged in domains such as personalized medicine and advertising to develop individualized treatment regimes. However, a critical deficit of existing algorithms is that they force a decision even when predictions are uncertain, a risky approach in high-stakes settings. The ability to abstain, that is, to defer to a safe default or an expert, is crucial but largely unexplored in this context. To remedy this, we introduce a framework for policy learning with abstention, in which policies that choose not to assign a treatment to some customers/patients receive a small, additive reward on top of the value of a random guess. We propose a two-stage learner that first identifies a set of near-optimal policies and then constructs an abstention class based on disagreements between the policies. We establish fast $O(1/n)$-type regret guarantees for the abstaining policy when propensities are known, and show how to extend these guarantees to the unknown-propensity case via a doubly robust (DR) objective. Furthermore, we demonstrate that our abstention framework is a versatile tool with direct applications to several other core problems in policy learning. We use our algorithm as a black box to obtain improved guarantees under margin conditions without the common realizability assumption. We also show that abstention provides a natural connection to both distributionally robust policy learning, where it acts as a hedge against small data shifts, and safe policy improvement, where the goal is to improve upon a baseline policy with high probability.
APA
Sawarni, A., Jin, J., Whitehouse, J. & Syrgkanis, V.. (2026). Policy Learning with Abstention . Proceedings of The 29th International Conference on Artificial Intelligence and Statistics, in Proceedings of Machine Learning Research 300:4465-4473 Available from https://proceedings.mlr.press/v300/sawarni26a.html.

Related Material