Supervised Metric Regularization Through Alternating Optimization for Multi-Regime Physics-Informed Neural Networks
Enzo Nicolás Spotorno, Josafat Leal Filho, & Antônio Augusto Fröhlich · 2026
Read Supervised Metric Regularization Through Alternating Optimization for Multi-Regime Physics-Informed Neural Networks by Enzo Nicolás Spotorno, Josafat Leal Filho, & Antônio Augusto Fröhlich (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-2602-09980.txt.
Description
Physics-informed neural networks learn solutions that obey known equations, but sharp regime changes, such as bifurcations, can make them blur distinct behaviors together. A 2026 paper led by Enzo Nicolas Spotorno, with two co-authors, proposes a topology-aware variant called TAPINN. It organizes the latent space so different regimes stay metrically apart.
A phase-based alternating schedule balances the metric and physics goals during training. On the Duffing oscillator, the method cuts the physics residual to about half the standard level, with lower gradient variance than a Sobolev-error baseline. It also uses about a fifth as many parameters as a hypernetwork alternative, avoiding memorization that violates physics.
Catalog record
| CREATOR | Enzo Nicolás Spotorno, Josafat Leal Filho, & Antônio Augusto Fröhlich |
|---|---|
| YEAR | 2026 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 13 KB |
| VERSION | 1.0 |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2602-09980