AWhat changed
Combined momentum with inexpensive approximate Newton–Schulz orthogonalization of matrix-valued hidden-layer updates, with empirical neural-network training evaluations.
Scientific scope
Combined momentum with inexpensive approximate Newton–Schulz orthogonalization of matrix-valued hidden-layer updates, with empirical neural-network training evaluations. Approximate finite matrix iterations are not exact orthogonalization or a global training guarantee. Other parameter groups use an auxiliary optimizer. Orthogonal-gradient, Shampoo and numerical-iteration predecessors retain credit; later MuonClip is distinct.
Evidence
The linked original sources describe the stated contribution and its evidence limits. WRS has not independently rerun the experiments or certified a proof. Approximate finite matrix iterations are not exact orthogonalization or a global training guarantee. Other parameter groups use an auxiliary optimizer. Orthogonal-gradient, Shampoo and numerical-iteration predecessors retain credit; later MuonClip is distinct.
Contributors
- Source-listed authors and technical contributors · SCIENCE: Source-listed authors or explicitly credited technical contributors for the cited versions. Family entries retain source-specific attribution; a combined list is not one original-paper byline or equal individual work. Names alone do not establish resolved identities.
- Muon: An optimizer for hidden layers in neural networks
- Muon official implementation and requested citation
Provisional WRS assessment, edition 2026-10-16.1. Independent expert calibration is pending. This grade does not certify correctness or assign ownership.
Full scientific record · Browse contribution grades · Request a correction