arXiv：cs.LG（机器学习，全量分类）· Ying Cao, Kun Yuan, Ali H. Sayed·· 1 天前AI 评分30
分布式对抗训练算法的逃逸效率研究
On the Escaping Efficiency of Distributed Adversarial Training Algorithms
AI 导读
研究对比了集中式与去中心化对抗训练算法在多智能体环境中的逃逸效率，发现当扰动边界足够小且批量较大时，去中心化策略（含共识与扩散）比集中式策略更快逃离局部极小值，倾向更平坦的极小值。但扰动边界增大后该趋势可能不再成立，仿真结果验证了理论发现并系统比较了两类算法的模型性能。
正文
Abstract:Adversarial training has been widely studied in recent years due to its role in improving model robustness against adversarial attacks. This paper focuses on comparing different distributed adversarial training algorithms--including centralized and decentralized strategies--within multi-agent learning environments. Previous studies have highlighted the importance of model flatness in determining robustness. To this end, we develop a general theoretical framework to study the escaping efficiency of these algorithms from local minima, which is closely related to the flatness of the resulting models. We show that when the perturbation bound is sufficiently small (i.e., when the attack strength is relatively mild) and a large batch size is used, decentralized adversarial training algorithms--including consensus and diffusion--are guaranteed to escape faster from local minima than the centralized strategy, thereby favoring flatter minima. However, as the perturbation bound increases, this trend may no longer hold. In the simulation results, we illustrate our theoretical findings and systematically compare the performance of models obtained through decentralized and centralized adversarial training algorithms. The results highlight the potential of decentralized strategies to enhance the robustness of models in distributed settings.
| Subjects: | Machine Learning (cs.LG) |
| Cite as: | arXiv:2509.11337 [cs.LG] |
| (or arXiv:2509.11337v2 [cs.LG] for this version) | |
| https://doi.org/10.48550/arXiv.2509.11337 arXiv-issued DOI via DataCite |
Submission history
From: Ying Cao [view email]
[v1]
Sun, 14 Sep 2025 16:28:20 UTC (1,568 KB)
[v2]
Thu, 1 Oct 2026 15:19:49 UTC (16,587 KB)
来源：arXiv：cs.LG（机器学习，全量分类） · arxiv.org