Data-dependent Exploration for Online Reinforcement Learning from Human Feedback

Fuente: arXiv
Saved in:
Bibliographic Details
Main Authors: Zhang, Zhen-Yu, Tang, Yuting, Zhang, Jiandong, Ma, Lanjihong, Sugiyama, Masashi
Format: Preprint
Published: 2026
Subjects:
Online Access:
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866910194270732288
author Zhang, Zhen-Yu
Tang, Yuting
Zhang, Jiandong
Ma, Lanjihong
Sugiyama, Masashi
author_facet Zhang, Zhen-Yu
Tang, Yuting
Zhang, Jiandong
Ma, Lanjihong
Sugiyama, Masashi
contents Online reinforcement learning from human feedback (RLHF) has emerged as a promising paradigm for aligning large language models (LLMs) by continuously collecting new preference feedback during training. A foundational challenge in this setting is exploration, which requires algorithms that enable the LLMs to generate informative comparisons that improve sample-efficiency in online RLHF. Existing exploration strategies often derive bonuses via on-policy expectations, which are difficult to estimate reliably from the limited historical preference data available during training; as a result, the policy can prematurely down-weight under-explored regions that may contain high-value behaviors. In this paper, we propose data-dependent exploration for preference optimization (DEPO), a simple and scalable method that leverages historical data to construct an extra uncertainty bonus for high-uncertainty regions, encouraging exploration toward potentially high-value data. Theoretically, we provide a data-dependent regret bound for the proposed algorithm, showing that it adapts to the hardness of the learning task itself and can be tighter than worst-case bounds in practice. Empirically, the proposed method consistently outperforms strong baselines across benchmarks, demonstrating improved sample efficiency.
format Preprint
id arxiv_https___arxiv_org_abs_2605_04477
institution arXiv
publishDate 2026
record_format arxiv
spellingShingle Data-dependent Exploration for Online Reinforcement Learning from Human Feedback
Zhang, Zhen-Yu
Tang, Yuting
Zhang, Jiandong
Ma, Lanjihong
Sugiyama, Masashi
Machine Learning
Online reinforcement learning from human feedback (RLHF) has emerged as a promising paradigm for aligning large language models (LLMs) by continuously collecting new preference feedback during training. A foundational challenge in this setting is exploration, which requires algorithms that enable the LLMs to generate informative comparisons that improve sample-efficiency in online RLHF. Existing exploration strategies often derive bonuses via on-policy expectations, which are difficult to estimate reliably from the limited historical preference data available during training; as a result, the policy can prematurely down-weight under-explored regions that may contain high-value behaviors. In this paper, we propose data-dependent exploration for preference optimization (DEPO), a simple and scalable method that leverages historical data to construct an extra uncertainty bonus for high-uncertainty regions, encouraging exploration toward potentially high-value data. Theoretically, we provide a data-dependent regret bound for the proposed algorithm, showing that it adapts to the hardness of the learning task itself and can be tighter than worst-case bounds in practice. Empirically, the proposed method consistently outperforms strong baselines across benchmarks, demonstrating improved sample efficiency.
title Data-dependent Exploration for Online Reinforcement Learning from Human Feedback
topic Machine Learning
url https://arxiv.org/abs/2605.04477