naPINN: Noise-Adaptive Physics-Informed Neural Networks for Recovering Physics from Corrupted Measurement
Hankyeol Kim, Pilsung Kang · 2026
Read naPINN: Noise-Adaptive Physics-Informed Neural Networks for Recovering Physics from Corrupted Measurement by Hankyeol Kim, Pilsung Kang (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-2602-02547.txt.
Description
Hankyeol Kim and 1 co-author make physics-informed neural networks survive corrupted measurements, in a 2026 arXiv paper. These networks solve inverse problems and recover governing equations by baking physical laws into training. Complex noise and gross outliers degrade them badly, and the noise distribution is usually unknown beforehand.
Their noise-adaptive network embeds an energy-based model inside the training loop to learn the distribution of prediction residuals. The learned energy landscape powers a trainable reliability gate that filters data points whose residuals look anomalous. A rejection cost prevents the trivial escape of discarding valid data wholesale. Benchmarks on partial differential equations with non-Gaussian noise and varying outlier rates followed. The method significantly beat existing robust baselines, isolating outliers while reconstructing the true dynamics under severe corruption.
Catalog record
| CREATOR | Hankyeol Kim, Pilsung Kang |
|---|---|
| YEAR | 2026 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 49 KB |
| VERSION | 1.0 |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2602-02547