← back
arXivXianliang Li, Zihan Zhang, Weiyang Liu, Han BaoTue, Jun 2, 2026, 9:54 AM PDT
score 16.4

Why momentum helps Muon optimizer in AI training

Original: Denoise First, Orthogonalize Later: Understanding Momentum in Muon via Spectral Filtering

Source: arxiv.org

Writing ELI5 summary…