arXivDaily arXiv每日学术速递 周一至周五更新
arXiv 2602.01581cs.LG
收藏

近优主动偏好学习及其在大语言模型对齐中的应用

Towards Optimality in Active Preference Learning

University of Arizona(亚利桑那大学);Department of Computer Science(计算机科学系)

Yao Zhao, Kwang-Sung Jun

AI总结:

本文提出两种主动学习算法,通过实例依赖标签复杂性保证和贪心方法提升大语言模型对齐的样本效率。

摘要:

对齐大型语言模型(LLMs)依赖于高质量的人类偏好标签数据集,这些数据集的收集成本很高。尽管已有研究通过主动学习提高样本效率,但许多现有方法仍采用传统实验设计标准,如G-或D-最优性。这些目标并未针对偏好学习的结构进行优化,留下设计问题特定算法的空间。在本工作中,我们识别出一个特定于偏好学习的简单直觉,质疑现有设计目标的适用性。受这一见解启发,我们提出了两种主动学习算法。第一种提供了该设置下的首个实例依赖标签复杂性保证,第二种是一个简单实用的贪心方法。我们在真实世界偏好数据集上评估了我们的算法,并观察到与现有方法相比样本效率的提升。

英文摘要:

Aligning large language models (LLMs) depends on high-quality datasets of human preference labels, which are costly to collect. Although active learning has been studied to improve sample efficiency relative to passive collection, many existing approaches adopt classical experimental design criteria such as G- or D-optimality. These objectives are not tailored to the structure of preference learning, leaving open the design of problem-specific algorithms. In this work, we identify a simple intuition specific to preference learning that calls into question the suitability of these existing design objectives. Motivated by this insight, we propose two active learning algorithms. The first provides the first instance-dependent label complexity guarantee for this setting, and the second is a simple, practical greedy method. We show the optimality of our instance-dependent term up to a constant depending on the logit bound, compared with a lower-bound characteristic. We evaluate our algorithms on real-world preference datasets and synthetic data observing improved sample efficiency compared to existing methods.

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。

↑