Certified and accurate computation of function space norms of deep neural networks
Johannes Gründler, Moritz Maibaum, Philipp Petersen · 2026
Read Certified and accurate computation of function space norms of deep neural networks by Johannes Gründler, Moritz Maibaum, Philipp Petersen (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-2603-06431.txt.
Description
Gründler and his team develop a certified method for norms of neural networks, needed for error control in PDE solving. Trained ones can be probed only at few points, and point values alone cannot yield sure bounds without strong assumptions. Beyond a black-box view, they exploit network structure, combining interval covers on axis-aligned boxes with adaptive tuning and sum aggregation. Box bounds on values and derivatives grow into global integral bounds, with closing theory for values, Jacobians, and Hessians proving Lebesgue and Sobolev norms.
Catalog record
| CREATOR | Johannes Gründler, Moritz Maibaum, Philipp Petersen |
|---|---|
| YEAR | 2026 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 111025 |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2603.06431