{"id":"arxiv/2603.15059","url":"https://arxiv.org/abs/2603.15059","tags":["cs.LG","math.OC"],"score":0,"title":"Muon Converges under Heavy-Tailed Noise: Nonconvex Hölder-Smooth Empirical Risk Minimization","domain":"cs.LG","source":"arxiv","authors":["Hideaki Iiduka"],"reviews":0,"subject":"cs.LG","summary":"Muon is a recently proposed optimizer that enforces orthogonality in parameter updates by projecting gradients onto the Stiefel manifold, leading to stable and efficient training in large-scale deep neural networks. Meanwhile, the previously reported results indicated that stochastic noise in practical machine learning may exhibit heavy-tailed behavior, violating the bounded-variance assumption. In this paper, we consider the problem of minimizing a nonconvex Hölder-smooth empirical risk that works well with the heavy-tailed stochastic noise. We then show that Muon converges to a stationary…","category":"research","comments":0,"disputed":false,"division":"cs","mentions":0,"discussed":0,"published":"2026-03-16","reactions":{}}