Accelerating Newton-Schulz Iteration for Orthogonalization via Chebyshev-type Polynomials
About
The problem of computing optimal orthogonal approximation to a given matrix has attracted growing interest in machine learning. Notable applications include the recent Muon optimizer or Riemannian optimization on the Stiefel manifold. Among existing approaches, the Newton-Schulz iteration has emerged as a particularly effective solution, as it relies solely on matrix multiplications and thus achieves high computational efficiency on GPU hardware. Despite its efficiency, the method has inherent limitations - its coefficients are fixed and thus not optimized for a given matrix. In this paper we address this issue by proposing a Chebyshev-optimized version of Newton-Schulz (CANS). Based on the Chebyshev's alternance theorem, we theoretically derive optimal coefficients for the 3-rd order Newton-Schulz iteration and apply a Remez algorithm to compute optimal higher-degree polynomials. We leverage these polynomials to construct controlled approximate orthogonalization schemes, which is of interest in deep learning applications. Practically, we demonstrate the method's effectiveness in two key applications: orthogonalization in the Muon optimizer, and providing an efficient retraction alternative for Riemannian optimization on the Stiefel manifold.
Related benchmarks
| Task | Dataset | Result | Rank | |
|---|---|---|---|---|
| Question Answering | ARC Easy (test) | Accuracy65.57 | 74 | |
| Language Modeling Evaluation | General Language Evaluation Suite AE, AC, SciQ, MMLU, MMLU-P, HS, OBQA, PIQA, RACE, WG, CSQA, AGI (test) | AE Score69.32 | 27 | |
| Matrix Orthogonalization | Synthetic Matrices 128 x 128 | Error1.311 | 5 | |
| Matrix Orthogonalization | Synthetic Matrices h=128 w=512 | Error1.309 | 5 | |
| Matrix Orthogonalization | Synthetic Matrices 128 x 1024 | Error1.311 | 5 |