Lundi 3 mars 2025 – LJLL : Claire Boyer (Université Paris-Saclay)

Titre: A primer on physics-informed machine learning Résumé Physics-informed machine learning combines the expressiveness of data-based approaches with the interpretability of physical models. In this context, we consider a general regression problem where the empirical risk is regularized by a partial differential equation that quantifies the physical inconsistency. Practitioners often…

plus

Lundi 3 février 2025 – LJLL : Maha Daoud (ENSTA Paris)

Titre : A class of parabolic fractional reaction-diffusion systems with control of total mass: Theory and numerics Résumé : In this talk based on [1, 2], we present some new results about global-in-time existence of strong solutions to a class of fractional parabolic reaction–diffusion systems posed in a bounded open…

plus

Lundi 20 janvier 2025 – Visio : Tabea Tscherpel (TU Darmstadt)

Titre: Energy consistent time discretisation of port-Hamiltonian systems Résumé Various ordinary and partial differential equations arising from physics can be written as port-Hamiltonian systems. Their Hamiltonian function represents an energy that is conserved or dissipated along solutions. Numerical schemes are energy consistent, if the Hamiltonian is preserved or dissipated also…

plus

Jeudi 19 décembre 2024 – LJLL : Zhao Junming (Harbin Institute of Technology, Chine)

Titre: Photon tunneling heat transfer in particulate system: physical characteristics and homogenization theory Résumé Radiative transfer equation (RTE) is the commonly accepted continuum scale governing equation for radiative heat transfer in particulate system. However, its applicability is questionable for non-random, densely and regularly packed particulate systems, due to dependent scattering…

plus

Lundi 9 décembre 2024 – LJLL : Alena Kopanicakova (Université de Toulouse)

Titre: Towards trustworthy use of scientific machine-learning in large-scale numerical simulations Résumé Recently, scientific machine learning (SciML) has expanded the capabilities of traditional numerical approaches by simplifying computational modeling and providing cost-effective surrogates. However, SciML models suffer from the absence of explicit error control, a computationally intensive training phase, and…

plus