The paper demonstrates that row-wise normalization can achieve higher population accuracy than full-batch Adam, a proxy for random-reshuffling Adam, and exact-SVD Muon in high-dimensional multiclass classification. Under an isotropic Gaussian-cloud data model, this advantage stems from row normalization’s class-wise Euclidean geometry preserving decision-boundary directions, whereas Adam’s coordinate-wise geometry and Muon’s spectral geometry introduce distortions. The advantage also persists in more general settings, including class means with independently oriented covariances, power-law spectra with class-mean exponent below one, and even under heavily anisotropic test noise, though it can reverse when covariances are diagonal and close, unless a random rotation aligns them with Adam’s coordinate axes.
arXiv:2606. 04662v1 Announce Type: cross Abstract: Muon improves training efficiency over Adam in large language-model training by about two times, but the local geometric source of this advantage remains unclear.
arXiv:2609.39114v1 Announce Type: new Abstract: This paper examines how row-wise renormalization affects Muon, focusing on the gap between NorMuon's worst-case guarantees and its practical performanc...
The paper investigates why the orthogonal optimiser Muon outperforms Adam in large language model pretraining by analysing the spectral properties of Transformer loss landscapes. It finds that Muon’s momentum buffers exhibit an anisotropic spectral profile with a volatile head and a tolerant bulk, enabling larger effective step sizes. Building on this insight, the authors propose Spectral‑Aware Muon (SAMuon) and a lightweight variant, which adjust the bulk scaling while keeping the head unchanged, achieving 13–24 % fewer training tokens than Muon without extra FLOPs.
arXiv:2610.07103v1 Announce Type: cross Abstract: Muon, an optimizer known for its efficiency, has a clear interpretation for matrix-valued updates, but convolutional kernels are stored as four-dimen...
arXiv:2602. 05725v3 Announce Type: replace Abstract: Muon updates matrix parameters via the matrix sign of the gradient and has shown strong empirical gains, yet its dynamics and scaling behavior remain unclear in theory.
arXiv:2606. 25971v1 Announce Type: new Abstract: Modern neural network training relies on optimizers such as Adam and Muon which act on each weight matrix as a single object.
The paper analyzes the stochastic optimizer Muon, which orthogonalizes an exponential moving average of minibatch gradients, and compares its behavior to the baseline SignSGD on a high‑dimensional matrix‑valued least squares problem. It derives deterministic dynamics showing Muon’s sensitivity to an effective batch size and its failure to resolve small eigenmodes when whitening is exact and the effective batch is small, while Momentum and finite Newton–Shulz iterations can mitigate this. The study presents a phase diagram in terms of data and target exponents, indicating when Muon or SignSGD is preferable based on covariance spectrum steepness and target loss scaling.
arXiv:2610.02705v1 Announce Type: cross Abstract: The Muon optimizer derives its update rule for hidden linear layers by solving a local linearization of the loss penalized by the spectral norm, moti...
arXiv:2606. 09658v1 Announce Type: cross Abstract: Muon has recently emerged as a state-of-the-art optimizer for pretraining Large Language Models (LLMs) and vision classifiers.
arXiv:2610.18294v1 Announce Type: cross Abstract: Releasing a neural network checkpoint can leak information about its training data, and existing reconstruction results rely on gradient-descent trai...
arXiv:2607. 13246v1 Announce Type: cross Abstract: Muon has recently emerged as a strong optimizer for large-scale deep learning, where it reshapes gradient updates through approximate orthogonalization and has been reported to outperform Adam and AdamW in large language model training.
Musec introduces MomentUm SpEctral Clipping, an optimizer-level, architecture‑agnostic technique that replaces Muon’s spectral flattening with selective spectral clipping to stabilize training. By clipping singular values above a threshold while preserving the momentum’s spectral structure, Musec addresses loss spikes and unbounded weight growth without requiring architecture‑specific changes. Soft Musec, an efficient implementation using smooth spectral saturation via coupled Newton‑Schulz iterations, offers convergence guarantees in nonconvex nonsmooth stochastic optimization and empirically improves stability across diverse learning rates and model sizes.