Gespeichert in:
Bibliographische Detailangaben
Hauptverfasser: Ryu, J. Jon, Kwon, Jeongyeol, Koppe, Benjamin, Jun, Kwang-Sung
Format: Preprint
Veröffentlicht: 2025
Schlagworte:
Online-Zugang:https://arxiv.org/abs/2502.10826
Tags: Tag hinzufügen
Keine Tags, Fügen Sie den ersten Tag hinzu!
_version_ 1866911053300891648
author Ryu, J. Jon
Kwon, Jeongyeol
Koppe, Benjamin
Jun, Kwang-Sung
author_facet Ryu, J. Jon
Kwon, Jeongyeol
Koppe, Benjamin
Jun, Kwang-Sung
contents We consider off-policy selection and learning in contextual bandits, where the learner aims to select or train a reward-maximizing policy using data collected by a fixed behavior policy. Our contribution is two-fold. First, we propose a novel off-policy selection method that leverages a new betting-based confidence bound applied to an inverse propensity weight sequence. Our theoretical analysis reveals that this method achieves a significantly improved, variance-adaptive guarantee over prior work. Second, we propose a novel and generic condition on the optimization objective for off-policy learning that strikes a different balance between bias and variance. One special case, which we call freezing, tends to induce low variance, which is preferred in small-data regimes. Our analysis shows that it matches the best existing guarantees. In our empirical study, our selection method outperforms existing methods, and freezing exhibits improved performance in small-sample regimes.
format Preprint
id arxiv_https___arxiv_org_abs_2502_10826
institution arXiv
publishDate 2025
record_format arxiv
spellingShingle Improved Offline Contextual Bandits with Second-Order Bounds: Betting and Freezing
Ryu, J. Jon
Kwon, Jeongyeol
Koppe, Benjamin
Jun, Kwang-Sung
Machine Learning
Information Theory
We consider off-policy selection and learning in contextual bandits, where the learner aims to select or train a reward-maximizing policy using data collected by a fixed behavior policy. Our contribution is two-fold. First, we propose a novel off-policy selection method that leverages a new betting-based confidence bound applied to an inverse propensity weight sequence. Our theoretical analysis reveals that this method achieves a significantly improved, variance-adaptive guarantee over prior work. Second, we propose a novel and generic condition on the optimization objective for off-policy learning that strikes a different balance between bias and variance. One special case, which we call freezing, tends to induce low variance, which is preferred in small-data regimes. Our analysis shows that it matches the best existing guarantees. In our empirical study, our selection method outperforms existing methods, and freezing exhibits improved performance in small-sample regimes.
title Improved Offline Contextual Bandits with Second-Order Bounds: Betting and Freezing
topic Machine Learning
Information Theory
url https://arxiv.org/abs/2502.10826