Saved in:
Bibliographic Details
Main Authors: Yu, Hanlin, Hauberg, Søren, Hartmann, Marcelo, Klami, Arto, Arvanitidis, Georgios
Format: Preprint
Published: 2025
Subjects:
Online Access:https://arxiv.org/abs/2505.24665
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866908958587879424
author Yu, Hanlin
Hauberg, Søren
Hartmann, Marcelo
Klami, Arto
Arvanitidis, Georgios
author_facet Yu, Hanlin
Hauberg, Søren
Hartmann, Marcelo
Klami, Arto
Arvanitidis, Georgios
contents Real world data often lie on low-dimensional Riemannian manifolds embedded in high-dimensional spaces. This motivates learning degenerate normalizing flows that map between the ambient space and a low-dimensional latent space. However, if the manifold has a non-trivial topology, it can never be correctly learned using a single flow. Instead multiple flows must be `glued together'. In this paper, we first propose the general training scheme for learning such a collection of flows, and secondly we develop the first numerical algorithms for computing geodesics on such manifolds. Empirically, we demonstrate that this leads to highly significant improvements in topology estimation.
format Preprint
id arxiv_https___arxiv_org_abs_2505_24665
institution arXiv
publishDate 2025
record_format arxiv
spellingShingle Learning Geometry and Topology via Multi-Chart Flows
Yu, Hanlin
Hauberg, Søren
Hartmann, Marcelo
Klami, Arto
Arvanitidis, Georgios
Machine Learning
Real world data often lie on low-dimensional Riemannian manifolds embedded in high-dimensional spaces. This motivates learning degenerate normalizing flows that map between the ambient space and a low-dimensional latent space. However, if the manifold has a non-trivial topology, it can never be correctly learned using a single flow. Instead multiple flows must be `glued together'. In this paper, we first propose the general training scheme for learning such a collection of flows, and secondly we develop the first numerical algorithms for computing geodesics on such manifolds. Empirically, we demonstrate that this leads to highly significant improvements in topology estimation.
title Learning Geometry and Topology via Multi-Chart Flows
topic Machine Learning
url https://arxiv.org/abs/2505.24665