Geometry-Preserving Neural Architectures on Manifolds with Boundary

Fuente: arXiv
Gespeichert in:
Bibliographische Detailangaben
Hauptverfasser: Elamvazhuthi, Karthik, Biswal, Shiba, Rosenblum, Kian, Katyal, Arushi, Qu, Tianli, Ma, Grady, Sonthalia, Rishi
Format: Preprint
Veröffentlicht: 2026
Schlagworte:
Online-Zugang:
Tags: Tag hinzufügen
Keine Tags, Fügen Sie den ersten Tag hinzu!
_version_ 1866917244372516864
author Elamvazhuthi, Karthik
Biswal, Shiba
Rosenblum, Kian
Katyal, Arushi
Qu, Tianli
Ma, Grady
Sonthalia, Rishi
author_facet Elamvazhuthi, Karthik
Biswal, Shiba
Rosenblum, Kian
Katyal, Arushi
Qu, Tianli
Ma, Grady
Sonthalia, Rishi
contents Preserving geometric structure is important in learning. We propose a unified class of geometry-aware architectures that interleave geometric updates between layers, where both projection layers and intrinsic exponential map updates arise as discretizations of projected dynamical systems on manifolds (with or without boundary). Within this framework, we establish universal approximation results for constrained neural ODEs. We also analyze architectures that enforce geometry only at the output, proving a separate universal approximation property that enables direct comparison to interleaved designs. When the constraint set is unknown, we learn projections via small-time heat-kernel limits, showing diffusion/flow-matching can be used as data-based projections. Experiments on dynamics over S^2 and SO(3), and diffusion on S^{d-1}-valued features demonstrate exact feasibility for analytic updates and strong performance for learned projections
format Preprint
id arxiv_https___arxiv_org_abs_2602_03082
institution arXiv
publishDate 2026
record_format arxiv
spellingShingle Geometry-Preserving Neural Architectures on Manifolds with Boundary
Elamvazhuthi, Karthik
Biswal, Shiba
Rosenblum, Kian
Katyal, Arushi
Qu, Tianli
Ma, Grady
Sonthalia, Rishi
Machine Learning
Systems and Control
Optimization and Control
Preserving geometric structure is important in learning. We propose a unified class of geometry-aware architectures that interleave geometric updates between layers, where both projection layers and intrinsic exponential map updates arise as discretizations of projected dynamical systems on manifolds (with or without boundary). Within this framework, we establish universal approximation results for constrained neural ODEs. We also analyze architectures that enforce geometry only at the output, proving a separate universal approximation property that enables direct comparison to interleaved designs. When the constraint set is unknown, we learn projections via small-time heat-kernel limits, showing diffusion/flow-matching can be used as data-based projections. Experiments on dynamics over S^2 and SO(3), and diffusion on S^{d-1}-valued features demonstrate exact feasibility for analytic updates and strong performance for learned projections
title Geometry-Preserving Neural Architectures on Manifolds with Boundary
topic Machine Learning
Systems and Control
Optimization and Control
url https://arxiv.org/abs/2602.03082