Actualités
Pour vous abonner à la newsletter Bayésienne et recevoir dans votre boîtes aux lettres les dernières actualités, vous pouvez suivre ce lien.
All About that... Seminar Series
Challenges in High Dimensional Bayesian Modelling
November 7, 2025. 14:00 - 17:00, IHP, Pierre Grisvard (room 314)
The seminar is open to everyone, but please confirm your attendance by registering at this link
Julyan Arbel (Inria, Université Grenoble Alpes) - Bayesian deep learning, overview and challenges
Abstract: Bayesian deep learning is appealing as it combines the coherence and natural uncertainty quantification of the Bayesian paradigm together with the expressivity and compositional flexibility of deep neural networks. It has its roots in pioneering work by Radford Neal and David Mackay in the 1990s on Bayesian neural networks. Its strengths lie in principled uncertainty quantification, improved data efficiency, and adaptability, making it impactful in safety-critical fields like healthcare and autonomous systems. In this talk I will provide an overview of Bayesian deep learning and discuss some of the key challenges the field faces in addressing modern machine learning problems.
Marion Naveau (Institut Agro Rennes-Angers) - High-dimensional variable selection in non-linear mixed effects models. Application in plant breeding
Abstract: The problem of variable selection in high-dimensional context, where the number of covariates exceeds the number of observations, is well studied in the context of standard regression models. However, few tools are currently available to address this issue for nonlinear mixed-effects models, where data are collected repeatedly across multiple individuals. My thesis focused on developing a high-dimensional variable selection procedure for these models, examining both its practical implementation and theoretical properties. This method is based on a Gaussian spike-and-slab prior and the SAEM algorithm (Stochastic Approximation of the Expectation-Maximization Algorithm). Its utility is illustrated through an application aimed at identifying genetic markers potentially involved in the senescence process of winter wheat. Furthermore, metamodeling approaches are being developed to reduce computation time when the regression function is costly to evaluate.
A third speaker will be announced soon...
Pour recevoir toutes les informations concernant ce séminaire, n'oubliez pas de vous inscrire à l'info lettre du groupe ou à celle dédiée au séminaire : s'abonner.