PRISM: Distribution-free Adaptive Computation of Matrix Functions for Accelerating Neural Network Training
Shenghao Yang 1,2 , Zhichao Wang 1,2 , Oleg Balabanov 1,2 , N. Benjamin Erichson 1,3 , and Michael W. Mahoney 1,2,3 1 International Computer Science Institute 2 University of California, Berkeley 3 Lawrence Berkeley National Laboratory {shenghao.yang, zhichao.wang, obalaban}@berkeley.edu erichson@ic · 2026
Read PRISM: Distribution-free Adaptive Computation of Matrix Functions for Accelerating Neural Network Training by Shenghao Yang 1,2 , Zhichao Wang 1,2 , Oleg Balabanov 1,2 , N. Benjamin Erichson 1,3 , and Michael W. Mahoney 1,2,3 1 International Computer Science Institute 2 University of California, Berkeley 3 Lawrence Berkeley National Laboratory {shenghao.yang, zhichao.wang, obalaban}@berkeley.edu erichson@ic (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-2601-22137.txt.
Description
Shenghao Yang and co-authors accelerate the matrix computations inside neural network training. Matrix functions such as square roots, inverse roots. And orthogonalization sit at the heart of preconditioned gradient methods, the optimizers that reshape gradients for faster learning. Iterative algorithms compute them using mostly matrix multiplications, which suits graphics processors.
The team's 2026 preprint presents PRISM, a distribution-free adaptive method for computing these matrix functions. Adaptive computation spends effort where the numbers demand it, rather than running a fixed schedule. Faster, stabler matrix functions would speed the training of large models that rely on advanced optimizers, cutting both time and energy per experiment.
Catalog record
| CREATOR | Shenghao Yang 1,2 , Zhichao Wang 1,2 , Oleg Balabanov 1,2 , N. Benjamin Erichson 1,3 , and Michael W. Mahoney 1,2,3 1 International Computer Science Institute 2 University of California, Berkeley 3 Lawrence Berkeley National Laboratory {shenghao.yang, zhichao.wang, obalaban}@berkeley.edu erichson@ic |
|---|---|
| YEAR | 2026 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 96 KB |
| VERSION | 1.0 |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2601-22137