arXivDaily arXiv每日学术速递 周一至周五更新
arXiv 2510.04686cs.LGcs.AI
收藏

优化器如何隐式地偏置模型合并的损失景观?

How does the optimizer implicitly bias the model merging loss landscape?

University of Luxembourg(卢森堡大学);ETH Zürich, MPI for Intelligent Systems(苏黎世联邦理工学院、智能系统研究所);Idiap Research Institute(Idiap研究机构);ELLIS Institute Tübingen, MPI for Intelligent Systems, Tübingen AI Center(图宾根ELLIS研究所、智能系统研究所、图宾根人工智能中心)

Chenxiang Zhang, Alexander Theus, Damien Teney, Antonio Orvieto, Jun Pang, Sjouke Mauw

AI总结:

本文研究了优化动态如何影响损失景观几何及其对模型合并成功的影响,发现有效噪声尺度统一了不同优化器组件对模型合并的影响,并揭示了学习率、权重衰减等参数对有效噪声尺度的调节作用。

摘要:

模型合并将具有不同能力的独立解决方案合并成一个,同时保持相同的推理成本。两种流行方法是线性插值,即简单平均多个模型权重,以及任务算术,即通过微调和基础模型的差异获得任务向量进行组合。尽管在实践中很有用,但使合并有效的性质仍不明确。本文探讨了优化动态如何影响损失景观几何及其对合并成功的影响。我们证明了一个单一量——有效噪声尺度——统一了不同优化器组件对模型合并的影响。在各种架构和数据集上,合并成功率是有效噪声尺度的非单调函数,具有一个独特的最优值。分解这一量,我们发现较大的学习率、更强的权重衰减、较小的批量大小和数据增强等独立调节有效噪声尺度,并表现出相同定性趋势。与先前将优化器噪声与单个极小值的平坦度或泛化能力联系起来的工作不同,我们表明它也影响全局损失景观,预测何时独立训练的解决方案可以成功合并。我们的发现扩展了对优化如何塑造损失景观几何及其对模型合并后果的理解,表明训练动态可能进一步被调节以提高模型合并。

英文摘要:

Model merging combines independent solutions with different capabilities into a single one while maintaining the same inference cost. Two popular approaches are linear interpolation, which simply averages multiple model weights, and task arithmetic, which combines task vectors obtained by the difference between finetuned and base models. While useful in practice, what properties make merging effective are poorly understood. This paper explores how the optimization dynamics affect the loss landscape geometry and its impact on merging success. We show that a single quantity -- the effective noise scale -- unifies the impact of different optimizer components on model merging. Across architectures and datasets, merging success is a non-monotonic function of the effective noise scale, with a distinct optimum. Decomposing this quantity, we find that larger learning rates, stronger weight decay, smaller batch sizes, and data augmentation all independently modulate the effective noise scale and exhibit the same qualitative trend. Unlike prior work connecting optimizer noise to the flatness or generalization of individual minima, we show that it also affects the global loss landscape, predicting when independently trained solutions can be successfully merged. Our findings broaden the understanding of how optimization shapes the loss landscape geometry and its consequences for model merging, suggesting that training dynamics could be further manipulated to improve model merging.

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。

↑