Foundation models for physics

About

Abstract

In this lecture, we will examine the emerging paradigm of foundation models for physics.

Rapid advances on scientific problems such as jet tagging in particle physics, gravitational-wave detection in astrophysics, or weather prediction in meteorology suggest that the next breakthroughs in the physical sciences will be driven by machine learning. Recent work has increasingly adopted the foundation model paradigm for science, in which models are trained on broad and heterogeneous datasets spanning multiple physical regimes, and later adapted to a wide range of downstream tasks, including prediction, parameter inference, and other inverse problems. We will show how fundamental concepts such as multiscale dynamics, partial observability, and generalization, guide the design of these foundation models, through examples and hands-on exercises in fluid dynamics and solar physics.

Link to google colab:

https://colab.research.google.com/drive/1v0cNw7TveOTGUDj71AcW908NwGX_-pQ7?usp=sharing

Bio

Rudy Morel is an Associate Research Scientist in the Scientific Computing Core at the Flatiron Institute in New York. Previously, he was a Research Fellow in the Center for Computational Mathematics and completed his PhD in Applied Mathematics at the École Normale Supérieure in Paris.

Please note that pictures and videos might be taken during the event. The pictures and videos might be used for communication about the event. By joining the lecture, you are agreeing to being featured in these communication actions.

00:00:00 Slide 1

00:01:42 Slide 2

00:06:33 Slide 3

00:08:15 Slide 4