2022/2023 PSL Week on Statistical Physics and Machine Learning

The aim of the Machine Learning - Statistical Physics intensive week is to present methods, ideas and connections between these two fields. In fact, methods and ideas developed in statistical physics of disordered systems can provide additional new tools to analyze the high-dimensional non-convex problems that emerge in machine learning.

Course Information

• Instructors: Francis Bach (INRIA & DI-ENS) and Giulio Biroli (LPENS).
• Dates: 04-08 March, 2024
• Location: ENS-Ulm, Physics Department.

Evaluation

Assiduity + Paper review.

Course Description

During the first part of the week, after a general introduction, we will focus on simple machine learning problems and analyze them by rigorous and exact (but non-rigorous) statistical physics method. This will be helpful to concrete present some of the connections between ML and statistical physics. In particular, we will introduce the replica method which has proven very useful in physics and other branches of science, as recognized by the 2021 Nobel Prize in Physics to Giorgio Parisi. The first part is meant to be « hands-on » and it will include problem classes.

The second part will be devoted to more realistic models and current research questions, such as the Double Descent phenomenon and Convex and Non-Convex Optimization.

Course Schedule

The schedule will consists in two sessions per day of 2 hours each from Monday to Friday (possibly ending Friday morning).