Muon Converges under Heavy-Tailed Noise: Nonconvex Hölder-Smooth Empirical Risk Minimization

Hideaki Iiduka [email protected] · 2026

Read Muon Converges under Heavy-Tailed Noise: Nonconvex Hölder-Smooth Empirical Risk Minimization by Hideaki Iiduka [email protected] (2026) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2603-15059.txt.

Description

Iiduka proves that the Muon optimizer still converges when training noise has heavy tails. Muon keeps updates orthogonal by projecting gradients onto the Stiefel manifold, which steadies large network training. Real noise can break the bounded-variance assumption that older proofs need. He studies nonconvex Holder-smooth empirical risk under a bound that allows heavy-tailed noise. Muon reaches a stationary point under that bound. He also shows it converges faster than mini-batch stochastic gradient descent.

Catalog record

CREATORHideaki Iiduka [email protected]
YEAR2026
COLLECTIONPAPERS
FORMATTXT
FILE SIZE83401
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2603.15059

Browse the full library · About the Tower of Babel