arXiv cs.AI· Zhanhong Jiang, Qisai Liu, Soumik Sarkar·· 12 小时前AI 评分22
观点论文：为何分布式 LLM 后训练与去中心化对齐如今不可或缺
Vision Paper: Why Distributed LLM Post-Training and Decentralized Alignment Are Now Essential
AI 导读
一篇观点论文主张分布式学习（含联邦与去中心化）应成为 LLM 后训练的一等范式，而非仅作隐私附加项。作者指出 RLHF 和 DPO 隐含“偏好数据应集中”的假设，并从隐私与数据主权、多元性与代表性公平、鲁棒性与可扩展性三方面论证其不成立。文章形式化了 reference drift、偏好异质性和对比损失的 DP-效用权衡等挑战，并提出算法方向与研究议程。
正文 · 原文
Abstract:This vision paper argues that distributed learning, including federated and decentralized, should become a first-class paradigm in LLM post-training, not merely a privacy add-on. Current methods like RLHF and DPO rest on an invisible assumption: preference data should be centralized. We contend this assumption is scientifically unjustified, practically unsustainable, and ethically problematic. As LLMs deploy into healthcare, law, finance, and multilingual communities, preference data originates from populations that cannot or will not centralize. We develop a three-pillar argument grounded in (1) privacy and data sovereignty, (2) pluralism and representational equity, and (3) robustness and scalability. We formalize key technical challenges unique to this intersection, reference drift, preference heterogeneity, and the DP-utility trade-off for contrastive losses, and propose concrete algorithmic directions, and sketch a research agenda. Our analysis shows distributed post-training is not only necessary but tractable, and ignoring it risks AI that is legally noncompliant, representationally narrow, and brittle at scale.
| Comments: | 10 pages, 4 figures, IEEE BigData 2026 |
| Subjects: | Artificial Intelligence (cs.AI) |
| Cite as: | arXiv:2611.05841 [cs.AI] |
| (or arXiv:2611.05841v1 [cs.AI] for this version) | |
| https://doi.org/10.48550/arXiv.2611.05841 arXiv-issued DOI via DataCite (pending registration) |
Submission history
From: Zhanhong Jiang [view email]
[v1]
Wed, 4 Nov 2026 02:30:51 UTC (314 KB)
来源：arXiv cs.AI · arxiv.org