“Generative models: Theory, learning and environmental applications” day

Day “Generative models: Theory, learning and environmental applications”

The Geolearning Chair and the RESSTE network organize a scientific day on the theme of recent methodological progress for generative models (GANs, Denoising Diffusion Models, etc.) and their use in the fields of climate, environment and more generally geosciences, with interventions from experts in theoretical aspects , methodological and applied.

Registrations are closed.

Program

9:50 a.m. – 10:00 a.m. Opening of the day

10:00 a.m.-12:00 p.m. Introduction(s) to generative models
Moderation: Thomas Opitz and Denis Allard

Francis Bach (SIERRA–Inria, ENS–PSL) An alternative view of denoising diffusion models [Presentation ]
Gerard Biau (LPSM, Sorbonne University) A primer on Generative Adversarial Networks [Presentation ]
Pascal Yiou (LSCE, CEA) A stochastic weather generator as an emulator for extreme climate events [Presentation ]

12:00 p.m. – 13:30 p.m. Buffet lunch on site

13:30 p.m.-14:30 p.m. Recent methodological and applied advances
Moderation: Thomas Romary

Pierre Marion (EPFL) Diffusion models for image generation: algorithms and (a bit of) theory [Presentation ]
Maxime Beauchamp (IMT Atlantique) Linking Geostatistics SPDE & Data Assimilation through deep learning for the generation of (large) ensembles in Geosciences [Presentation ]

14:30 p.m.-15:30 p.m. Exploration of new methods and implementations (Part I)
Moderation: Liliane Bel

Ferdinand Bhavsar (Mines-Paris PSL) Categorical Denoising Diﬀusion Models using Markov Jump Processes [Presentation ]
David Metivier (MISTEA, INRAE) Deep Learning for lake pollution forecast [Presentation ]

15:30 p.m.-16:00 p.m. Coffee break

16:00-17:00 p.m. Exploration of new methods and implementations (Part II)
Moderation: Valérie Monbet

Lucia Clarotto, Hugo Gangloff (MIA Paris Saclay, AgroParisTech/INRAE) Physics-informed neural networks for Matérn-SPDEs: attempts and challenges [Presentation ]
Yuke Xie (Mines-Paris PSL) Deep Generative Models for Stochastic Geophysical Imaging and Uncertainty Quantification [Presentation ]

Francis Bach presents generative diffusion models

Gérard Biau presents GANs

Events

Conference on stochastic weather generators

The Geolearning Chair is contributing to the organization of the conference on Stochastic Weather Generators, which will be held in Grenoble from December 2nd to 4th, 2025. This international workshop aims to explore cutting-edge methods and emerging challenges for simulating weather variables: precipitation, temperatures, etc.

Events

Two presentations of the Geolearning chair at JDS 2025

Mike Pereira and Thomas Romary presented their work at the last Statistics Days, organized by the SFDS.

Events

Geolearning in the spotlight at the Spatial Statistics conference

Researchers from the Geolearning Chair will be honored at the next international Spatial Statistics conference, which will be held from July 15 to 18, 2025, in the Netherlands.