探索 Delta 学习假说的极限：DPO 中单对生成器带来的偏差
Exploring the Limits of the Delta Learning Hypothesis: Single-Pair Generator Biases in Direct Preference Optimization
AI 导读
一项 arXiv 研究检验了 delta-learning 假说在 Direct Preference Optimization（DPO）中的局限，该假说主张以一对大小模型中较大模型的回答作为偏好标签来生成训练数据。
正文 · 原文
Abstract:Direct Preference Optimization (DPO) typically relies on preference pairs generated from diverse model pools and ranked by an external LLM judge. The recently proposed delta-learning hypothesis suggests that comparable supervision can instead be obtained from a single pair of models by treating the larger model's response as preferred. We investigate the limits of this approach across model families and scales. We show, both theoretically and empirically, that fixed model pairs introduce systematic, generator-dependent biases that are predictive of the preference label independent of response quality. Consequently, DPO trained on single-pair data consistently underperforms multi-model preference data, despite the larger generator producing higher-quality responses, and we trace this gap to training dynamics consistent with reward hacking on these biases. We further show that simply increasing the number of generators, even without an external judge, is enough to partially mitigate this effect. Our results highlight generator diversity as a critical factor in preference-data construction.
| Subjects: | Machine Learning (cs.LG); Artificial Intelligence (cs.AI); Computation and Language (cs.CL) |
| Cite as: | arXiv:2610.37058 [cs.LG] |
| (or arXiv:2610.37058v1 [cs.LG] for this version) | |
| https://doi.org/10.48550/arXiv.2610.37058 arXiv-issued DOI via DataCite (pending registration) |
Submission history
From: Giuseppe Curci [view email]
[v1]
Tue, 27 Oct 2026 11:24:50 UTC (4,289 KB)
来源：arXiv cs.AI · arxiv.org