Réunion

Bayesian inference for inverse problems

Annonce

Traditionally, inverse problems are written as the minimization of a criterion combining data attachment and a regularization term. Just as traditionally, the Bayesian approach to the question represents these terms in the form of likelihood and prior distributions, which together allow the formation of a poster distribution to be maximized. The well-known advantage of this second approach is that it allows the quantification of uncertainties associated with estimates, which is an asset in terms of the explainability of the process.
This thematic day offers an opportunity for discussion on this topic, covering both its more traditional aspects (MCMC approach and variations) and more modern aspects, such as variational inference and plug-and-play priors.

The topics covered by the call for contributions are open in terms of the problem support (signal, image, or other) and applications. They may, without limitation, cover the following topics:

• variational Bayesian approaches
• plug-and-play algorithms at the interface between Bayesian approaches and machine learning
• approaches for quantifying uncertainty (in high dimensions)
• scaling algorithms to high dimensions (observations and parameters)
• multimodal estimators and distributions

Contributions from young researchers are particularly welcome.

Keynote speakers:

• Julien Stoehr (Ceremade, Paris Dauphine – PSL)
• Mame Diarra Fall (Univ. de Rouen Normandie)

To submit a proposal for contribution, please send a one-page abstract (title and summary) before April 11, 2026, at:

• jean-baptiste.courbot – at – uha.fr
• pierre-antoine.thouvenin – at – centralelille.fr

—

When : 11 june 2026
Where : IHP, Paris, Amphi Choquet-Bruhat
Organization : Jean-Baptiste Courbot (IRIMAS, Mulhouse) et Pierre-Antoine Thouvenin (Cristal, Lille)