Equivariant Flow Matching on Homogeneous Spaces

Flow matching enables the training of networks that can flow samples from an initial to a target distribution. The network is regressed against simple vector fields that are conditioned on end points in the target distribution. In the Euclidean case, these conditional vector fields induce flows along straight lines. Various generalisations have been proposed to perform flow matching on spaces where straight lines are not well-defined, such as Riemannian manifolds and Lie groups. Riemannian flow matching applies to many domains, but typically requires expensive simulation, while flow matching on Lie groups is simulation-free but only works on Lie groups. In this work, we develop equivariant flow matching on homogeneous spaces, which generalises flow matching on Lie groups, while remaining simulation-free. We also derive all generators connecting pairs of position-orientations, generalising, and perform a similar derivation for event-velocity space, the Lorentzian analogue of position-orientation space, so that we can perform flow matching on these spaces.

Authors

Publication Details

Journal
arXiv (Cornell University)
Published
2026-11-17
DOI
https://doi.org/10.48550/arxiv.2611.22492
Primary Topic
Model Reduction and Neural Networks
Type
preprint
Controls
|||
ALL TIME
JAN
FEB
MAR
APR
MAY
JUN
JUL
AUG
SEP
OCT
NOV
preprint

Equivariant Flow Matching on Homogeneous Spaces

Gijs Bellaard, Finn M. Sherry
arXiv (Cornell University)
Model Reduction and Neural Networks
preprint

Equivariant Flow Matching on Homogeneous Spaces

Gijs Bellaard, Finn M. Sherry
preprint en

Abstract

Flow matching enables the training of networks that can flow samples from an initial to a target distribution. The network is regressed against simple vector fields that are conditioned on end points in the target distribution. In the Euclidean case, these conditional vector fields induce flows along straight lines. Various generalisations have been proposed to perform flow matching on spaces where straight lines are not well-defined, such as Riemannian manifolds and Lie groups. Riemannian flow matching applies to many domains, but typically requires expensive simulation, while flow matching on Lie groups is simulation-free but only works on Lie groups. In this work, we develop equivariant flow matching on homogeneous spaces, which generalises flow matching on Lie groups, while remaining simulation-free. We also derive all generators connecting pairs of position-orientations, generalising, and perform a similar derivation for event-velocity space, the Lorentzian analogue of position-orientation space, so that we can perform flow matching on these spaces.

arXiv (Cornell University)
Model Reduction and Neural Networks
AI Navigator

Ask Laika to Summarize, Analyze, and Connect papers live on the map.

Summarize Papers & Methodologies

Extract key findings, datasets, and comparative methods across publications.

Benchmark Rankings & Visual Analytics

Rank top research institutions, authors, funders, topics, and journals by Field-Weighted Citation Impact (FWCI) and paper volume with instant charts.

Connect Distant Disciplines

Bridge topological clusters on the map to find hidden collaborative intersections.

Laika access coming soon