Bias-Constrained Diffusion Schedules for PDE Emulations: Reconstruction Error Minimization and Efficient Unrolled Training
Constantin Le Cleï, Nils Thuerey, Xiaoxiang Zhu · 2026
Read Bias-Constrained Diffusion Schedules for PDE Emulations: Reconstruction Error Minimization and Efficient Unrolled Training by Constantin Le Cleï, Nils Thuerey, Xiaoxiang Zhu (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-2604-08357.txt.
Description
Le Cleï, Thuerey, and Zhu improve diffusion models that stand in for fluid and flow equations over time. Standard noise plans, they show, leave needless rebuilding error and make long training runs costly. Their adaptive plan caps the bias from training exposure to cut that error, and a cheaper proxy training pass steadies long forecasts. Tests on forced flow, wave, and transonic cases beat diffusion and deterministic rivals on short accuracy and long stability.
Catalog record
| CREATOR | Constantin Le Cleï, Nils Thuerey, Xiaoxiang Zhu |
|---|---|
| YEAR | 2026 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 65679 |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2604.08357