Geometry-Preserving Neural Architectures on Manifolds with Boundary
Karthik Elamvazhuthi, Shiba Biswal, Kian Rosenblum, Arushi Katyal, Tianli Qu, Grady Ma, Rishi Sonthalia · 2026
Read Geometry-Preserving Neural Architectures on Manifolds with Boundary by Karthik Elamvazhuthi, Shiba Biswal, Kian Rosenblum, Arushi Katyal, Tianli Qu, Grady Ma, Rishi Sonthalia (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-2602-03082.txt.
Description
Neural networks that must respect geometric constraints are organized by Karthik Elamvazhuthi and 6 co-authors in a 2026 paper. Architectures enforce such constraints through projections, exponential-map updates, constrained output layers, or manifold differential equations. The authors unify them by where and how enforcement happens, either throughout intermediate layers or only at the final output of the network.
That perspective reveals gaps in existing theory. The authors prove approximation theorems for several architecture families on prox-regular constraint sets, including smooth manifolds with boundary. Experiments cover synthetic dynamics on spheres, disks, and rotation groups, plus real protein backbone data. Final augmentation proved simpler and won most tasks. When the constraint set is unknown, projections can be learned from diffusion models.
Catalog record
| CREATOR | Karthik Elamvazhuthi, Shiba Biswal, Kian Rosenblum, Arushi Katyal, Tianli Qu, Grady Ma, Rishi Sonthalia |
|---|---|
| YEAR | 2026 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 88 KB |
| VERSION | 1.0 |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2602-03082