Geometric Generalization of Neural Operators from Kernel Integral Perspective
Mingyu Han, Daniel Zhengyu Huang, Yuhan Wang, Yanshu Zhang, Jiayi Zhou · 2026
Read Geometric Generalization of Neural Operators from Kernel Integral Perspective by Mingyu Han, Daniel Zhengyu Huang, Yuhan Wang, Yanshu Zhang, Jiayi Zhou (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-2602-01498.txt.
Description
Mingyu Han and co-authors study geometric generalization in neural operators via kernel integrals. They mimic PDE solution maps for many queries. Variable shapes make transfer to unseen shapes central in engineering design. Their kernel-integral view shows how linear models extend geometrically, linking model learning to fast kernel sums. They propose a multiscale point-cloud model from Ewald sums, joining Fourier terms with local geometry corrections. Guarantees follow for singular-kernel models, and tests show robust transfer on varied shapes, Laplace-type maps, and a nonlinear vehicle flow case.
Catalog record
| CREATOR | Mingyu Han, Daniel Zhengyu Huang, Yuhan Wang, Yanshu Zhang, Jiayi Zhou |
|---|---|
| YEAR | 2026 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 95506 |
| LICENSE | http://creativecommons.org/licenses/by-sa/4.0/ |
| RIGHTS | CC BY SA |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2602.01498