Domain decomposition of large neural network surrogate models
T. Gödde, E. H. Atzema, B. Rosić, [0.8em] University of Twente, Faculty of Engineering Technology, Applied Mechanics and Data Analysis, PO Box 217, 7500AE, Enschede, the Netherlands Tata Steel Nederland Technology B.V., PO Box 10000, NL-1970 CA IJmuiden, the Netherlands Vienna University of Technology, E307-04-01 Digital Engineering group, Institute of Engineering Design and Product Development, Lehargasse 6, 1060 Vienna Correspondence: [email protected] · 2026
Read Domain decomposition of large neural network surrogate models by T. Gödde, E. H. Atzema, B. Rosić, [0.8em] University of Twente, Faculty of Engineering Technology, Applied Mechanics and Data Analysis, PO Box 217, 7500AE, Enschede, the Netherlands Tata Steel Nederland Technology B.V., PO Box 10000, NL-1970 CA IJmuiden, the Netherlands Vienna University of Technology, E307-04-01 Digital Engineering group, Institute of Engineering Design and Product Development, Lehargasse 6, 1060 Vienna Correspondence: [email protected] (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-2603-26396.txt.
Description
Gödde, Atzema, and Rosić split large neural network surrogates into local models over subdomains. A single network can miss sharp local bends unless it grows very large. Their domain decomposition trains a smaller network on each part of the input space in parallel. Extra terms in the local losses keep the pieces in agreement at the seams. They enforce the match with plain and augmented Lagrange multiplier methods, and test both on synthetic compression problems in two and three dimensions.
Catalog record
| CREATOR | T. Gödde, E. H. Atzema, B. Rosić, [0.8em] University of Twente, Faculty of Engineering Technology, Applied Mechanics and Data Analysis, PO Box 217, 7500AE, Enschede, the Netherlands Tata Steel Nederland Technology B.V., PO Box 10000, NL-1970 CA IJmuiden, the Netherlands Vienna University of Technology, E307-04-01 Digital Engineering group, Institute of Engineering Design and Product Development, Lehargasse 6, 1060 Vienna Correspondence: [email protected] |
|---|---|
| YEAR | 2026 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 83974 |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2603.26396