Research map / Learning theory and algorithms / Matrix and tensor methods
The Polar Express: Optimal Matrix Sign Methods and their Application to the Muon Algorithm
Noah Amsel, David Persson, Christopher Musco, Robert M. Gower
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| replace Jordan's polynomial iteration with Polar Express inside Muon | Muon · FineWeb 1B tokens (GPT-2-Small, no weight decay) | validation loss: 3.639 -> 3.588 | Improved |
| replace You's polynomial iteration with Polar Express inside Muon | Muon · FineWeb 1B tokens (GPT-2-Small, no weight decay) | validation loss: 3.629 -> 3.588 | Improved |
| use Muon with Polar Express instead of AdamW | GPT-2-Small · FineWeb 1B tokens (GPT-2-Small, no weight decay) | validation loss: 4.197 -> 3.588 | Improved |
| use Polar Express with 5 or 6 iterations instead of 2 or 3 iterations in Muon (ablation) | Muon + Polar Express · GPT-2-Small FineWeb | final validation loss | Improved |
| compute the polar factor exactly via SVD instead of 5 iterations (ablation) | Muon · GPT-2-Small FineWeb | final validation loss | No clear effect |
| compute the polar factor exactly via SVD instead of Polar Express (ablation) | Muon · GPT-2-Small FineWeb | runtime per step | Hurt |
| add weight decay 0.1 (ablation) | Muon + Polar Express · GPT-2 FineWeb | validation loss | No clear effect |
| extend training from 1 billion to 10 billion tokens (ablation) | Muon + Polar Express · GPT-2-Large FineWeb (weight decay) | validation loss | No clear effect |
Closest papers by meaning
- Phases of Muon: When Muon Eclipses SignSGD (NeurIPS 2026 · similarity 0.87)
- Towards Understanding the Power and Limits of the Muon Optimizer: A River-Valley Perspective (NeurIPS 2026 · similarity 0.86)
- On the Convergence Analysis of Muon (NeurIPS 2026 · similarity 0.86)
- From SGD to Muon: Adaptive Optimization via Schatten-p Norms (NeurIPS 2026 · similarity 0.86)
- When and Why SignSGD Outperforms SGD: A Theoretical Study Based on $\ell_1$-norm Lower Bounds (NeurIPS 2026 · similarity 0.85)
- Muon Dynamics as a Spectral Wasserstein Flow (NeurIPS 2026 · similarity 0.85)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
