mlsanityData to 14 Oct 2026

Training Flow Matching: The Role of Weighting and Parameterization

arXiv 2603.06454 ↗First posted 6 Mar 2026cs.CVcited by 4

On the map: Diffusion models → Flow matching, time series and science

Abstract (arXiv)

We study the training objectives of denoising-based generative models, with a particular focus on loss weighting and output parameterization, including noise-, clean image-, and velocity-based formulations. Through a systematic numerical study, we analyze how these training choices interact with the intrinsic dimensionality of the data manifold, model architecture, and dataset size. Our experiments span synthetic datasets with controlled geometry as well as image data, and compare training objectives using quantitative metrics for denoising accuracy (PSNR across noise levels) and generative quality (FID). Rather than proposing a new method, our goal is to disentangle the various factors that matter when training a flow matching model, in order to provide practical insights on design choices.

Key facts. Cited by 4 papers in the map (data through 14 Oct 2026). Placed in Flow matching, time series and science, one of 14 directions in Diffusion models. Builds on 18 of its 26 references that are in the map. First posted on arXiv 6 Mar 2026.

Builds on

18 of 26 references.

Closest ancestors · 3

Same direction, newest first

  1. 2023Improving and generalizing flow-based generative models with minibatch optimal transport1 Feb
  2. 2022Flow Matching for Generative Modeling6 Oct
  3. 2022Flow Straight and Fast: Learning to Generate and Transfer Data with Rectified Flow7 Sep

Other references · 15

  1. 2020An Image is Worth 16x16 Words: Transformers for Image Recognition at Scalecited by 30,186
  2. 2020Denoising Diffusion Probabilistic Modelscited by 21,076
  3. 2015U-Net: Convolutional Networks for Biomedical Image Segmentationcited by 19,713
  4. 2020Score-Based Generative Modeling through Stochastic Differential Equationscited by 9,122
  5. 2015Deep Unsupervised Learning using Nonequilibrium Thermodynamicscited by 7,058
  6. 2022Building Normalizing Flows with Stochastic Interpolantscited by 995
  7. 2023Simple diffusion: End-to-end diffusion for high resolution imagescited by 422
  8. 2023Understanding Diffusion Objectives as the ELBO with Simple Data Augmentationcited by 304
  9. 2022Perception Prioritized Training of Diffusion Modelscited by 303
  10. 2023Efficient Diffusion Training via Min-SNR Weighting Strategycited by 246
  11. 2022Convergence of denoising diffusion models under the manifold hypothesiscited by 242
  12. 2015From Facial Parts Responses to Face Detection: A Deep Learning Approachcited by 42
3 more
  1. 2023Unmasking Bias in Diffusion Model Trainingcited by 13
  2. 2026Revisiting Diffusion Model Predictions Through Dimensionalitycited by 7
  3. 2025Loss Functions in Diffusion Models: A Comparative Studycited by 3

Built on by

Cited by year (2026 to 14 Oct)

Citing papers per year: 2026: 4.

4 papers.

  1. 2026FluxFlow: Conservative Flow-Matching for Astronomical Image Super-Resolutioncited by 2
  2. 2026JLT: Clean-Latent Prediction in Latent Diffusion Transformerscited by 1
  3. 2026Follow the Mean: Reference-Guided Flow Matchingcited by 0
  4. 2026Beyond the Manifold Hypothesis: Hybrid Spectral Parameterizations for Flow Matchingcited by 0

Similar, not cited

Close in wording, no citation either way

  1. 2025Carr'e du champ flow matching: better quality-generalisation tradeoff in generative models
  2. 2025On the Closed-Form of Flow Matching: Generalization Does Not Arise from Target Stochasticity
  3. 2025The Generation Phases of Flow Matching: a Denoising Perspective
  4. 2025A Theoretical Analysis of Discrete Flow Matching Generative Models
  5. 2025Flow Matching in the Low-Noise Regime: Pathologies and a Contrastive Remedy
  6. 2023Flow Matching in Latent Space
  7. 2026Flowing with Confidence
  8. 2023Multisample Flow Matching: Straightening Flows with Minibatch Couplings