arXivDaily arXiv每日学术速递 周一至周五更新
arXiv周末暂无论文更新,休息一下吧,周末愉快~~
arXiv 2511.04456cs.LG
收藏

重尾噪声下的联邦随机极小极大优化

Federated Stochastic Minimax Optimization under Heavy-Tailed Noises

Temple University(特拉华大学)

Xinwen Zhang, Hongchang Gao

AI总结:

本文针对联邦学习中重尾噪声下的非凸-PL极小极大优化问题,提出Fed-NSGDA-M和FedMuon-DA两种算法,首次在重尾噪声下取得严格理论收敛保证并通过实验验证。

摘要:

重尾噪声在非凸随机优化中受到越来越多的关注,因为大量实证研究表明,它比标准的有界方差假设提供了更现实的假设。本文研究了联邦学习中重尾梯度噪声下的非凸-PL极小极大优化问题。我们提出了两种新算法:Fed-NSGDA-M,它集成了归一化梯度;以及FedMuon-DA,它利用Muon优化器进行局部更新。这两种算法都旨在较温和的条件下有效解决联邦极小极大优化中的重尾噪声问题。我们从理论上证明,两种算法都达到了$O({1}/{(TNp)^{\ rac{s-1}{2s}}})$的收敛速率。据我们所知,这是首批在重尾噪声下具有严格理论保证的联邦极小极大优化算法。大量实验进一步验证了它们的有效性。

英文摘要:

Heavy-tailed noise has attracted growing attention in nonconvex stochastic optimization, as numerous empirical studies suggest it offers a more realistic assumption than standard bounded variance assumption. In this work, we investigate nonconvex-PL minimax optimization under heavy-tailed gradient noise in federated learning. We propose two novel algorithms: Fed-NSGDA-M, which integrates normalized gradients, and FedMuon-DA, which leverages the Muon optimizer for local updates. Both algorithms are designed to effectively address heavy-tailed noise in federated minimax optimization, under a milder condition. We theoretically establish that both algorithms achieve a convergence rate of $O({1}/{(TNp)^{\frac{s-1}{2s}}})$. To the best of our knowledge, these are the first federated minimax optimization algorithms with rigorous theoretical guarantees under heavy-tailed noise. Extensive experiments further validate their effectiveness.

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。

↑