重新定义对抗扰动机制，提升模型泛化能力
Fix the Loss, Not the Radius: Rethinking the Adversarial Perturbation of Sharpness-Aware Minimization
- 用损失预算替代固定半径，更契合曲率优化本质
- 在多个基准上超越SAM及变体，达当前最优性能
- 适合追求高泛化性的深度学习研究者参考
Sharpness-Aware Minimization (SAM) 通过在参数空间固定半径邻域内最小化最坏情况损失来提升泛化能力。现有SAM及其变体主要依赖一阶线性近似，而平坦极小值本质上是二阶（曲率）概念。本文重新审视这一不匹配问题，提出损失等价SAM（LE-SAM），将传统SAM中固定扰动半径的机制改为固定损失空间预算，有效消除梯度幅值主导的学习信号，使优化过程转向曲率主导项。在多种基准和任务上的广泛实验表明，LE-SAM展现出强大的泛化能力，持续优于SAM及其变体，达到当前最优性能。
原文摘要 · Abstract (English)
Sharpness-Aware Minimization (SAM) improves generalization by minimizing the worst-case loss within a fixed parameter-space radius neighborhood. SAM and its variants mainly rely on a first-order linearized surrogate, while flat minima are inherently a second-order (curvature) notion.We revisit this mismatch and propose Loss-Equated SAM (LE-SAM), which inverts the traditional SAM mechanism that fixed perturbation radius with a fixed loss-space budget,effectively removing gradient-norm-dominated learning signals and shifting optimization toward curvature-dominated terms. Extensive experiments across diverse benchmarks and tasks demonstrate the strong generalization ability of LESAM that consistently outperforms SAM and even its variants, achieving the state-of-the-art performance.
Thank you to arXiv for use of its open access interoperability. PaperDance 不是 arXiv 官方产品；中文卡片由大模型生成，请以原文为准。