EUVIMEDEUVIMED entdecken
Crossref

Stable generative modelling using Schrödinger bridges

Georg A Gottwald iD, Fengyi Li, Youssef Marzouk, Sebastian Reich iD

Philosophical Transactions of the Royal Society A: Mathematical, Physical and Engineering Sciences · 19.06.2025 · 383 (2299)

DOI: 10.1098/rsta.2024.0332 ↗

Kurz gefasst

Fachartikel in der Zeitschrift Philosophical Transactions of the Royal Society A: Mathematical, Physical and Engineering Sciences, online erschienen am 19.06.2025. Verfasst von Georg A Gottwald und 3 weiteren Autor:innen. Beteiligte Einrichtung: School of Mathematics and Statistics, University of Sydney, Sydney, New South Wales, Australia. Die Arbeit wurde bisher 4-mal zitiert (Crossref) und nennt 61 Literaturangaben. Gefördert von: US Deparment of Energy, Office of Advanced Scientific Computing Research, Deutsche Forschungsgemeinschaft, Australian Research Council.

AbstractSchlagwörterBibliografische AngabenZitierenVerwandte Studien

Abstract (Originalsprache)

We consider the problem of sampling from an unknown distribution for which only a sufficiently large number of training samples are available. Such settings have recently drawn considerable interest in the context of generative modelling and Bayesian inference. In this paper, we propose a generative model combining Schrödinger bridges and Langevin dynamics. Schrödinger bridges over an appropriate reversible reference process are used to approximate the conditional transition probability from the available training samples, which is then implemented in a discrete-time reversible Langevin sampler to generate new samples. By setting the kernel bandwidth in the reference process to match the time step size used in the unadjusted Langevin algorithm, our method effectively circumvents any stability issues typically associated with the time stepping of stiff stochastic differential equations. Moreover, we introduce a novel split-step scheme, ensuring that the generated samples remain within the convex hull of the training samples. Our framework can be naturally extended to generate conditional samples and to Bayesian inference problems. We demonstrate the performance of our proposed scheme through experiments on synthetic datasets, on a stochastic subgrid-scale parametrization conditional sampling problem, and on generating sample trajectories from a dynamical system using conditional sampling. This article is part of the theme issue ‘Generative modelling meets Bayesian inference: a new paradigm for inverse problems’.

Schlagwörter

Stichwörter der Autor:innen

Bayesian inferenceLangevin dynamicsSchrödinger bridgesgenerative modelling

Autor:innen und Institutionen

  • School of Mathematics and Statistics, University of Sydney, Sydney, New South Wales, Australia.
  • Massachusetts Institute of Technology, Cambridge, MA, USA.
  • Institute of Mathematics, University of Potsdam, Potsdam, Germany.

ORCID-Profile: Georg A Gottwald, Sebastian Reich

Förderung

  • US Deparment of Energy, Office of Advanced Scientific Computing Research
  • Deutsche Forschungsgemeinschaft
  • Australian Research Council

Zitieren

APA

Gottwald, G. A., Li, F., Marzouk, Y., & Reich, S. (2025). Stable generative modelling using Schrödinger bridges. Philosophical Transactions of the Royal Society A: Mathematical, Physical and Engineering Sciences, 383(2299). https://doi.org/10.1098/rsta.2024.0332

Vancouver

Gottwald GA, Li F, Marzouk Y, Reich S. Stable generative modelling using Schrödinger bridges. Philosophical Transactions of the Royal Society A: Mathematical, Physical and Engineering Sciences. 2025;383(2299). doi:10.1098/rsta.2024.0332

BibTeX
@article{Gottwald2025,
  author = {Gottwald, Georg A and Li, Fengyi and Marzouk, Youssef and Reich, Sebastian},
  title = {Stable generative modelling using Schrödinger bridges},
  journal = {Philosophical Transactions of the Royal Society A: Mathematical, Physical and Engineering Sciences},
  year = {2025},
  volume = {383},
  number = {2299},
  doi = {10.1098/rsta.2024.0332}
}
Kostenlos starten →