This one-day workshop aims to provide a broad overview of the use of surrogate models (or metamodels) in industrial applications such as electrical systems. These models are used to approximate complex simulators or physical and numerical systems with simpler ones.
The program will include a mix of industrial presentations showcasing practical use cases, and methodological talks presenting recent state-of-the-art approaches. The day will conclude with a poster session.
This workshop is part of the INRIA-EDF partnership: Gérer les systèmes électriques de demain.
| Time | Speaker | Title |
|---|---|---|
| 09:15 – 09:30 | — | Updates on the "Défi INRIA-EDF" |
| 09:30 – 10:15 | Alejandro Ribes | Hybrid reduced-order and learning-based metamodels of electrical machines: applications to digital twins and industrial inverse problems |
| 10:15 – 11:00 | Pierre-Louis Ruhlmann | Flow Matching for Robust Simulation-Based Inference under Model Misspecification |
| Coffee break | ||
| 11:15 – 12:00 | Manon Verbockhaven | Expressivity bottlenecks in neural networks within the framework of neural architecture search |
| Lunch break | ||
| 13:30 – 14:15 | Julien Pelamatti | tba |
| 14:15 – 15:00 | Patrick Gallinari | Neural Surrogates for Parametric PDEs: Learning to Extrapolate |
| Coffee break | ||
| 15:15 – 16:00 | Paul Strang | Model-based reinforcement learning for exact combinatorial optimization |
| 16:00 – 17:00 | Clément Dauvilliers | A self-supervised framework and architecture for geospatial data with an application to tropical cyclones |
| 17:00 – 18:00 | Poster Session | |
