NS-RGS uses Newton-Schulz iterations to avoid costly matrix decompositions in Riemannian optimization for orthogonal group synchronization, proving linear convergence with spectral initialization and showing 2x practical speedup.
Muon: An optimizer for hidden layers in neural networks
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
fields
stat.ML 1years
2026 1verdicts
UNVERDICTED 1representative citing papers
citing papers explorer
-
NS-RGS: Newton-Schulz based Riemannian gradient method for orthogonal group synchronization
NS-RGS uses Newton-Schulz iterations to avoid costly matrix decompositions in Riemannian optimization for orthogonal group synchronization, proving linear convergence with spectral initialization and showing 2x practical speedup.