Some remarks on the connection between generative diffusion and optimal transport

Abstract

We discuss three simple connections between generative diffusion and optimal transport. First, the joint law of the data and of their Ornstein-Uhlenbeck noising is, at every time, the solution of an entropic optimal-transport problem, at a temperature set by the noise level. Backward diffusion can consequently be read as an annealing which gradually decreases the entropic regularization. Second, we study the speciation transition--how the modes of the data distribution emerge during generation--and compare how it happens in diffusion and flow matching with respect to optimal transport. Using Gaussian mixtures as high-dimensional example, we show that the dynamical instability through which diffusion and flow matching separate the modes becomes, for optimal transport, an exponentially thin boundary layer at the beginning of generation, which is rounded by entropic regularization. Third, considering a generic forward dynamics with a general confining potential, we show that the backward process driven by the score also satisfies a drift-based variational principle whose minimal cost is the entropy difference between initial and final distributions.

One-Step Generative Modeling via Unbalanced Optimal Transport
Yirong Shen, Mengfei Xia, Junpeng Jing +2
arXiv26 Sep 2026PreprintarXiv licenceFull textGeneral mathematicsarXiv
Barron Optimal Transport I: Generative Modeling
Evan Dogariu, Joan Bruna
arXiv7 Oct 2026PreprintCC BY 4.0Full textAnalysis & PDEsarXiv
Remarks on a multi-dimensional transport equation with nonlocal velocity
Wanwan Zhang
arXiv3 Oct 2026PreprintarXiv licenceFull textAnalysis & PDEsarXiv
Connection between MP and DPP for Mean-Field Stochastic Optimal Control Problems
Guangzhi Cui, Tianyang Nie, Guangchen Wang
arXiv13 Oct 2026PreprintarXiv licenceFull textOptimisation & controlarXiv
Generative SDEs with Trainable Diffusion and Density-Based Losses
Angxiu Ni
arXiv30 Oct 2026PreprintarXiv licenceFull textDynamical systemsarXiv

Cite

BibTeX RIS

BibTeX

@misc{biroli2026some,
  title = {{Some remarks on the connection between generative diffusion and optimal transport}},
  author = {Giulio Biroli and Marc Mézard},
  howpublished = {arXiv},
  year = {2026},
  month = oct,
  doi = {10.48550/arxiv.2610.13614},
  eprint = {2610.13614},
  archivePrefix = {arXiv},
  url = {https://arxiv.org/abs/2610.13614},
}

RIS

TY  - UNPB
TI  - Some remarks on the connection between generative diffusion and optimal transport
AU  - Giulio Biroli
AU  - Marc Mézard
PY  - 2026
DA  - 2026/10/20
PB  - arXiv
DO  - 10.48550/arxiv.2610.13614
UR  - https://arxiv.org/abs/2610.13614
AB  - We discuss three simple connections between generative diffusion and optimal transport. First, the joint law of the data and of their Ornstein-Uhlenbeck noising is, at every time, the solution of an entropic optimal-transport problem, at a temperature set by the noise level. Backward diffusion can consequently be read as an annealing which gradually decreases the entropic regularization. Second, we study the speciation transition--how the modes of the data distribution emerge during generation--and compare how it happens in diffusion and flow matching with respect to optimal transport. Using Gaussian mixtures as high-dimensional example, we show that the dynamical instability through which diffusion and flow matching separate the modes becomes, for optimal transport, an exponentially thin boundary layer at the beginning of generation, which is rounded by entropic regularization. Third, considering a generic forward dynamics with a general confining potential, we show that the backward process driven by the score also satisfies a drift-based variational principle whose minimal cost is the entropy difference between initial and final distributions.
ER  -

Abstract

Related

Cite