arXivDaily arXiv每日学术速递 周一至周五更新
arXiv周末暂无论文更新,休息一下吧,周末愉快~~
arXiv 2605.04477cs.LG
收藏

基于数据的探索:面向人类反馈的在线强化学习

Data-dependent Exploration for Online Reinforcement Learning from Human Feedback

Center for Advanced Intelligence Project, RIKEN(日本理化学研究所先进智能研究中心);Graduate School of Frontier Sciences, The University of Tokyo(东京大学前沿科学研究生院);Northeastern University(东北大学);Zhejiang Gongshang University(浙江工商大学)

Zhen-Yu Zhang, Yuting Tang, Jiandong Zhang, Lanjihong Ma, Masashi Sugiyama

AI总结:

提出数据依赖探索的偏好优化方法(DEPO),利用历史数据构建不确定性奖励,鼓励探索高价值区域,理论证明其数据依赖的遗憾界更紧,实验表明样本效率提升。

摘要:

在线人类反馈强化学习(RLHF)通过在训练过程中持续收集新的偏好反馈,已成为对齐大型语言模型(LLM)的一种有前景的范式。该设置中的一个基本挑战是探索,这需要算法使LLM能够生成信息丰富的比较,从而提高在线RLHF中的样本效率。现有的探索策略通常通过策略内期望推导奖励,这些期望难以从训练期间有限的可用历史偏好数据中可靠估计;因此,策略可能会过早地降低对可能包含高价值行为的未充分探索区域的权重。在本文中,我们提出了数据依赖探索的偏好优化(DEPO),这是一种简单且可扩展的方法,利用历史数据为高不确定性区域构建额外的奖励,鼓励向潜在高价值数据探索。理论上,我们为所提算法提供了数据依赖的遗憾界,表明它能够适应学习任务本身的难度,并且在实践中比最坏情况界更紧。实验上,所提方法在多个基准测试中持续优于强基线,展示了改进的样本效率。

英文摘要:

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.

机构、总结、中文摘要均由AI分析而来,可能存在错误,请以论文正文为准。

↑