Explanation of Dynamic Physical Field Predictions using WassersteinGrad: Application to Autoregressive Weather Forecasting
Younes Essafouri [email protected], Laure Raynaud [email protected], Luciano Drozda [email protected], Laurent Risser [email protected] · 2026
Read Explanation of Dynamic Physical Field Predictions using WassersteinGrad: Application to Autoregressive Weather Forecasting by Younes Essafouri [email protected], Laure Raynaud [email protected], Luciano Drozda [email protected], Laurent Risser [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-2604-22580.txt.
Description
Essafouri and three colleagues tackle accounts of network forecasts on dynamic physical fields, as in autoregressive weather forecasting. Explaining forecasts moved from curiosity to a strict need in high-stakes settings. Gradient-based feature attribution is widely used for high-dimensional inputs. On images, SmoothGrad firms up accounts by averaging attribution maps from noised inputs, and the authors aim to adapt it to dynamic fields. Their finding pinpoints a failure mode in averaging perturbed attribution maps on dynamic fields: stochastic perturbations do not induce steady amplitude noise in the maps.
Catalog record
| CREATOR | Younes Essafouri [email protected], Laure Raynaud [email protected], Luciano Drozda [email protected], Laurent Risser [email protected] |
|---|---|
| YEAR | 2026 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 78089 |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2604.22580