"Physics-Informed Machine Learning" by Jordan Frécon-Deloire

Seminar by Jordan Frécon-Deloire

Abstract

This presentation will elucidate the core principles of physics-informed machine learning. The first part will focus on physics-informed neural networks (PINNs), detailing the methodology of modeling and training neural networks to efficiently approximate solutions to partial differential equations (PDEs) by incorporating physical knowledge. The second section will address the development of surrogates for a family of PDE solutions where, for instance, the initial conditions or PDE coefficients vary. Within this context, I will explore how neural operators, originally devised for mapping between two infinite-dimensional functions, can be informed by the underlying physics.

This seminar will be held in English.