Scrutinizing Variables for Checkpoint Using Automatic Differentiation
Xiang Fu, Xin Huang, Wubiao Xu, Shiman Meng, Weiping Zhang, Luanzheng Guo, Kento Sato · 2026
Read Scrutinizing Variables for Checkpoint Using Automatic Differentiation by Xiang Fu, Xin Huang, Wubiao Xu, Shiman Meng, Weiping Zhang, Luanzheng Guo, Kento Sato (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-2602-16010.txt.
Description
Huang and colleagues trim checkpoint costs in high performance computing by asking a sharper question: which data actually affects the output. Checkpoint restart saves program state on a schedule and eats storage and compute, yet much of what gets saved never influences the result, so it need not be saved at all.
Their method uses automatic differentiation to test every element inside each variable, marking pieces as critical or uncritical by whether they touch the output. Visual maps of those regions follow the physics and logic of each code, a useful check in itself. Across eight NAS Parallel Benchmark codes, dropping uncritical elements cuts checkpoint storage by up to twenty percent, a direct saving for long running simulations.
Catalog record
| CREATOR | Xiang Fu, Xin Huang, Wubiao Xu, Shiman Meng, Weiping Zhang, Luanzheng Guo, Kento Sato |
|---|---|
| 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/2602-16010