RECORD
Pullback Flow Matching on Data Manifolds
Machine Learning · Submitted 6 Oct 2024
abstract compiled · 2026-09-23 · no language-model review
The authors' main claim We demonstrate PFM's effectiveness through applications in synthetic data, protein dynamics and protein sequence data, generating novel proteins with specific properties.
The compiled record
The paper's own sentences, arranged. Each label is drawn from a closed vocabulary; the number beside it is the probability the classifier gave that label. Numbers in the tables are the authors', checked by code. No sentence in this record was written by a model.
Open the compiled record
The abstract, sentence by sentence
We propose Pullback Flow Matching (PFM), a novel framework for generative modeling on data manifolds. Unlike existing methods that assume or learn restrictive closed-form manifold mappings for training Riemannian Flow Matching (RFM) models, PFM leverages pullback geometry and isometric learning to preserve the underlying manifold's geometry while enabling efficient generation and precise interpolation in latent space. This approach not only facilitates closed-form mappings on the data manifold but also allows for designable latent spaces, using assumed metrics on both data and latent manifolds. By enhancing isometric learning through Neural ODEs and proposing a scalable training objective, we achieve a latent space more suitable for interpolation, leading to improved manifold learning and generative performance. We demonstrate PFM's effectiveness through applications in synthetic data, protein dynamics and protein sequence data, generating novel proteins with specific properties. This method shows strong potential for drug discovery and materials science, where generating novel samples with specific properties is of great interest.
background contribution method result interpretation main claim
| # | sentence | role | the authors' own work | states the main conclusion | strength |
|---|---|---|---|---|---|
| s0 | We propose Pullback Flow Matching (PFM), a novel framework for generative modeling on data manifolds. | contribution 1.00 | 0.89 | 0.11 | asserted 0.99 |
| s1 | Unlike existing methods that assume or learn restrictive closed-form manifold mappings for training Riemannian Flow Matching (RFM) models, PFM leverages pullback geometry and isometric learning to preserve the underlying manifold's geometry while enabling efficient generation and precise interpolation in latent space. | contribution 0.57 | 0.38 | 0.16 | asserted 1.00 |
| s2 | This approach not only facilitates closed-form mappings on the data manifold but also allows for designable latent spaces, using assumed metrics on both data and latent manifolds. | method 0.69 | 0.57 | 0.13 | asserted 0.97 |
| s3 | By enhancing isometric learning through Neural ODEs and proposing a scalable training objective, we achieve a latent space more suitable for interpolation, leading to improved manifold learning and generative performance. | method 0.56 | 0.92 | 0.22 | asserted 1.00 |
| s4 | We demonstrate PFM's effectiveness through applications in synthetic data, protein dynamics and protein sequence data, generating novel proteins with specific properties. | result 0.83 | 0.95 | 0.31 | asserted 1.00 |
| s5 | This method shows strong potential for drug discovery and materials science, where generating novel samples with specific properties is of great interest. | interpretation 0.91 | 0.34 | 0.22 | hedged possibility 0.56 |
What was examined, and what was not
- arXiv listing abstract v2
- full text
- figures and tables
- references and citing sentences
- relations to other papers
- P abstract
- S sections
- T body sentences
- C citations
- R cross-paper relations
The paper
Original abstract
We propose Pullback Flow Matching (PFM), a novel framework for generative modeling on data manifolds. Unlike existing methods that assume or learn restrictive closed-form manifold mappings for training Riemannian Flow Matching (RFM) models, PFM leverages pullback geometry and isometric learning to preserve the underlying manifold's geometry while enabling efficient generation and precise interpolation in latent space. This approach not only facilitates closed-form mappings on the data manifold but also allows for designable latent spaces, using assumed metrics on both data and latent manifolds. By enhancing isometric learning through Neural ODEs and proposing a scalable training objective, we achieve a latent space more suitable for interpolation, leading to improved manifold learning and generative performance. We demonstrate PFM's effectiveness through applications in synthetic data, protein dynamics and protein sequence data, generating novel proteins with specific properties. This method shows strong potential for drug discovery and materials science, where generating novel samples with specific properties is of great interest.
Human review
References
How the literature uses it
2 arXiv papers cite this in 2 sentences
Papers citing it · 1 sentence without a confident label
Cites it as background 1 paper
-
There are many applications in machine learning relying on the machinery of Riemannian geometry, especially in the context of flow matching [ref] or single-cell genomics [ref] .
Learning Lineage-guided Geodesics with Finsler Geometry · Riemannian and Finsler Metrics · 0.95
What it cites · 14 sentences without a confident label
Contests 1 paper
-
Low-dimensional approaches, such as latent diffusion [ref] or latent flow-matching [ref] , often overlook the geometric structure of the data, leading to inaccuracies in tasks requiring a faithful representation of the underlying manifold.
Flow Matching in Latent Space · Introduction · contest 0.75
Uses the method or tool of 1 paper
-
To evaluate our generative methods, we use the $1$-NN accuracy [ref] , which measures how well the generated point clouds match the reference point clouds.
Revisiting Classifier Two-Sample Tests · Generation Experiments · 1.00
Compares with 1 paper
-
This objective relies solely on a distance measure on the data manifold, simplifying the training process compared to [ref] while maintaining geometric fidelity. * We demonstrate our methods' effectiveness through experiments on synthetic data, high-dimensional molecular dynamics data, and experimental peptide sequences.
Pulling back symmetric Riemannian geometry for data analysis · Introduction · 0.83
Cites as support 1 paper
-
The primary objectives of learning isometries are i) to map the data manifold $( 𝒟, (·,·)^𝒟 )$ into a low-dimensional geodesic subspace of $(ℳ,(·,·)^ℳ )$, specifically $ℳ_d' ⊂ℳ$, and ii) to preserve local isometry, as motivated by Proposition 2.1 and Theorems 3.4, 3.6, and 3.8 from [ref] .
Pulling back symmetric Riemannian geometry for data analysis · Learning Objective · 0.98
Cites as background 1 paper
-
Since the rise of machine learning in the scientific domain, researchers have focused on developing larger models trained on increasingly massive datasets, as in weather forecasting [ref] and protein structure prediction [ref] .
A Foundation Model for the Earth System · Introduction · 0.99
Discussion
Know something about this paper? Leave a comment, or highlight any passage above to comment on it.