PRISM: Structured Optimization via Anisotropic Spectral Shaping
Yujie Yang · 2026
Read PRISM: Structured Optimization via Anisotropic Spectral Shaping by Yujie Yang (2026) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2602-03096.txt.
Description
Yujie Yang's solo 2026 paper proposes a new optimizer for training neural networks. Called PRISM, it enhances first-order spectral descent methods such as Muon with partial second-order information. Second-order information describes curvature, how sharply the loss changes, which can guide better steps but usually costs too much memory and computation for large models during training.
PRISM builds an efficient low-rank approximation to that curvature information through an innovation-augmented polar decomposition. It performs anisotropic spectral shaping, suppressing updates in high-variance directions while preserving strength where the signal dominates. The overhead is minimal, with no additional memory over first-order baselines. The paper presents PRISM as a practical way to bring curvature-adaptive behavior into spectral optimization.
Catalog record
| CREATOR | Yujie Yang |
|---|---|
| YEAR | 2026 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 28 KB |
| VERSION | 1.0 |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2602-03096