arXivDaily arXiv每日学术速递 周一至周五更新
arXiv 2603.03099cs.LGcs.AI
收藏

为何Adam能胜过SGD:二阶矩归一化产生更尖锐的尾部

Why Adam Can Beat SGD: Second-Moment Normalization Yields Sharper Tails

Department of Electrical and Computer Engineering, The Ohio State University(俄亥俄州立大学电气与计算机工程系);School of Electrical, Computer and Energy Engineering, Arizona State University(亚利桑那州立大学电气、计算机与能源工程学院)

Ruinan Jin, Yingbin Liang, Shaofeng Zou

AI总结:

本文揭示了Adam中的关键二阶矩归一化机制,并通过停止时间/鞅分析,在经典有界方差模型下,证明了Adam在高概率收敛行为上优于SGD,前者对置信参数δ的依赖为δ^{-1/2},而SGD则至少为δ^{-1}。

摘要:

尽管Adam在许多应用中表现出比SGD更快的实证收敛速度,但现有的大多数理论保证与SGD几乎相同,无法充分解释实证性能差距。在本文中,我们揭示了Adam中的关键二阶矩归一化,并开发了一种停止时间/鞅分析,该分析在经典有界方差模型(一个二阶矩假设)下,能够证明Adam在高概率收敛行为上优于SGD。具体而言,我们建立了两种方法高概率收敛行为之间的第一个理论区分:Adam对置信参数δ的依赖为δ^{-1/2},而SGD对应的高概率保证至少需要δ^{-1}的依赖。

英文摘要:

Despite Adam demonstrating faster empirical convergence than SGD in many applications, much of the existing theory yields guarantees essentially comparable to those of SGD, leaving the empirical performance gap insufficiently explained. In this paper, we uncover a key second-moment normalization in Adam and develop a stopping-time/martingale analysis that provably distinguishes Adam from SGD under the classical bounded variance model (a second moment assumption). In particular, we establish the first theoretical separation between the high-probability convergence behaviors of the two methods: Adam achieves a $δ^{-1/2}$ dependence on the confidence parameter $δ$, whereas corresponding high-probability guarantee for SGD necessarily incurs at least a $δ^{-1}$ dependence.

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。

↑