Organisateurs : David Rousseau (CNRS/Université Paris-Saclay, rousseau@ijclab.in2p3.fr), Marco Saitta (Sorbonne Université, marco.saitta@sorbonne-universite.fr) |
Division Astrophysique, division Champs et Particules et division Matière Condensée
|
Horaires : lundi 16h45 - jeudi 8h30 Salle Claudine Hermann Session posters : mercredi 18h30
|
Depuis quelques années, les approches basées sur l’intelligence artificielle (IA) sont de plus en plus développées et utilisées dans toutes les branches de la physique. De nombreuses initiatives (projets scientifiques, consortia, fédérations, GDR, instituts de recherche dédiés), de type top-down comme bottom-up, émergent partout en France. Ce mini-colloque a comme objectif de présenter un panorama de ces méthodes en physique, selon quatre axes principaux : l’astrophysique, la physique des particules, la physique des matériaux, la physique des systèmes complexes. D’autre part, la physique statistique permet de mieux comprendre comment un réseau de neurones "apprend"". In recent years, approaches based on artificial intelligence (AI) have been increasingly developed and used in all branches of physics. Numerous initiatives (scientific projects, consortia, federations, GDRs, dedicated research institutes), both top-down and bottom-up, are emerging all over France. The aim of this mini-colloquium is to present an overview of these methods in physics, focusing on four main areas: astrophysics, particle physics, materials physics and the physics of complex systems. Most AI methods are relevant. While image analysis is at the heart of AI, innovative neural network architectures are being developed to process data with complex structures derived from measurements or physics models. Heavy calculations can be accelerated by delegating part of them to lightweight surrogate models. The control of complex equipment can be inspired by techniques developed for autonomous vehicles. And our scientific papers must always be able to convince our peers, despite or thanks to the growing role of AI. On the other hand, statistical physics provides a better understanding of how a neural network "learns". Given the breadth of both the disciplines and the methods involved, this mini-colloquium toook place over 2 slots of 2h each, for oral presentations (invited and contributed), as well as poster sessions. |