Enhanced accuracy through ensembling of randomly initialized auto-regressive models for time-dependent PDEs
Ishan Khurjekar a, Indrashish Saha a, Lori Graham-Brady a, Somdatta Goswami a · 2025
Read Enhanced accuracy through ensembling of randomly initialized auto-regressive models for time-dependent PDEs by Ishan Khurjekar a, Indrashish Saha a, Lori Graham-Brady a, Somdatta Goswami a (2025) 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-2507-03863.txt.
Description
Ishan Khurjekar and co-authors use ensembles to steady machine learning models that forecast physics over time. Their 2025 paper deals with surrogate models for partial differential equations, which run far faster than classic solvers. A flaw is that step-by-step forecasts pile up error, so long predictions drift off course.
Their fix trains several copies of a model in parallel, each starting from random weights. At prediction time, the copies vote as a group, which damps the spread of small errors. The method was tested on three systems: stress in mixed materials, a reaction-diffusion pattern model, and a planetary shallow water model. Errors grew more slowly in every case, using only a few input steps.
Catalog record
| CREATOR | Ishan Khurjekar a, Indrashish Saha a, Lori Graham-Brady a, Somdatta Goswami a |
|---|---|
| YEAR | 2025 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 42 KB |
| VERSION | 1.0 |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2507-03863