当共享知识有害：模型融合中的谱过积累
When Shared Knowledge Hurts: Spectral Over-Accumulation in Model Merging
AI总结：
本文研究了模型融合中共享知识过积累的问题，提出SVC方法通过校准奇异值来恢复谱平衡，提升了模型融合和任务算术的性能。
摘要：
模型融合通过将多个微调模型的权重更新相加，提供了一种轻量级的替代方法，而非重新训练。现有方法主要针对解决任务更新之间的冲突，未处理共享知识过积累的失败模式。我们发现当任务共享对齐的谱方向（即重叠的奇异向量）时，简单的线性组合会反复积累这些方向，导致奇异值膨胀并使融合模型偏向共享子空间。为缓解此问题，我们提出Singular Value Calibration (SVC)，一种无需训练和数据的后处理方法，量化子空间重叠并重新缩放膨胀的奇异值以恢复平衡的谱。在视觉和语言基准上，SVC一致改进了强大的融合基线并实现了最先进的性能。此外，仅通过修改奇异值，SVC将任务算术的性能提高了13.0%。代码可在https://github.com/lyymuwu/SVC获取。
英文摘要：
Model merging combines multiple fine-tuned models into a single model by adding their weight updates, providing a lightweight alternative to retraining. Existing methods primarily target resolving conflicts between task updates, leaving the failure mode of over-counting shared knowledge unaddressed. We show that when tasks share aligned spectral directions (i.e., overlapping singular vectors), a simple linear combination repeatedly accumulates these directions, inflating the singular values and biasing the merged model toward shared subspaces. To mitigate this issue, we propose Singular Value Calibration (SVC), a training-free and data-free post-processing method that quantifies subspace overlap and rescales inflated singular values to restore a balanced spectrum. Across vision and language benchmarks, SVC consistently improves strong merging baselines and achieves state-of-the-art performance. Furthermore, by modifying only the singular values, SVC improves the performance of Task Arithmetic by 13.0%. Code is available at https://github.com/lyymuwu/SVC.
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。
