Description
Date depot: 18 janvier 2023
Titre: Physics Based Deep Learning - Learning PDE solvers
Directeur de thèse:
Patrick GALLINARI (ISIR (EDITE))
Encadrant :
Nicolas BASKIOTIS (ISIR (EDITE))
Domaine scientifique: Sciences et technologies de l'information et de la communication
Thématique CNRS : Intelligence artificielle
Resumé: Partial differential equations are the work-horse for the modeling of dynamic systems in many physical fields like thermodynamics, fluid dynamics, etc. Solving PDEs is computationally expensive and often prohibitive for complex or large systems. Developing fast methods for approximating PDE solutions or solving PDEs has motivated several research directions. Using ML to learn solvers is thus being explored for a few years now. The objective of the thesis is to explore new alternatives to these methods bypassing their limitations and that can handle multiple situations. Two main directions will be explored: (i) Learning solvers for differential equations and (ii) Reduced order and surrogate models.
Doctorant.e: Le Boudec Lise