跳到正文
arXiv cs.AI· Shangge Liu, Danyue Qian, Qi Fan, Yinghuan Shi, Wenbin Li, Lei Wang, Yang Gao, Dacheng Tao·· 4 小时前AI 评分38

模型合并通用代理正则化的谱框架研究

A Spectral Framework for Universal Proxy Regularization in Model Merging

AI 导读

arXiv 论文提出通用代理正则化的谱框架,用谱覆盖 α 与泄漏 β 刻画代理分布与待合并任务分布的关系,并证明更小的 (α, β) 通过更好的谱对齐带来更低的参数干扰。

正文 · 原文

View PDF HTML (experimental)

Abstract:Model merging constructs multi-task models without expensive joint training but often suffers from parameter interference. Some solutions mitigate this by constraining finetuning using data from the other tasks to be merged, violating the core independence and privacy assumption of model merging. Recent work offers preliminary empirical evidence that a universal, task-agnostic proxy distribution can substitute for such data, but leaves open when and why it suffices. To this end, we propose a \textit{Spectral Framework for Universal Proxy Regularization}, which characterizes the relationship between a proxy distribution and the distribution of other tasks to be merged through spectral coverage ($\alpha$) and leakage ($\beta$). Based on the framework, we derive a unified parameter interference bound: smaller $(\alpha, \beta)$ provably yields lower parameter interference through better spectral alignment. To exploit such a proxy, we augment the training objective with a proxy regularizer. Evaluating this regularizer during fine-tuning requires a concrete mechanism. We study three instantiations, including (J) an exact evaluation via Jacobian-vector products, (F) a frozen curvature approximation using the Fisher information matrix, and (O) an output-drift formulation that constrains the model's functional deviation on the proxy directly. We prove that all three coincide under the NTK linearization, where existing analyses of such regularizers stop, and go further to characterize standard non-linear fine-tuning, bounding how far each instantiation departs from its linearized reference and obtain a three-way hierarchy, zero for (J), proxy-independent for (F), and proxy-quality-scaling for (O). Empirical validation on the commonly used vision-language model CLIP confirms these theoretical predictions.
Subjects: Artificial Intelligence (cs.AI)
Cite as: arXiv:2611.17728 [cs.AI]
  (or arXiv:2611.17728v1 [cs.AI] for this version)
  https://doi.org/10.48550/arXiv.2611.17728

arXiv-issued DOI via DataCite (pending registration)

Submission history

From: Shangge Liu [view email]
[v1] Fri, 13 Nov 2026 04:16:40 UTC (3,491 KB)

来源:arXiv cs.AI · arxiv.org