This Hausdorff School aims to present, in four mini-courses, some of the recent mathematical developments in the field of statistical mechanics of spin glasses, in particular in view of its applications to the mathematical understanding of the functioning of machine learning. While this connection dates back to the 1980s, many spin glass-inspired methods have been applied to machine learning problems in recent years, often pushing the boundaries of the state-of-the-art. The main topics to be covered in the school are Hebbian neural networks and their generalisations, learning dynamics and non-convex optimisation on high-dimensional landscapes by stochastic gradient descent. All of these are related to the problem of understanding the topological complexity of high-dimensional random landscapes.
This school offers a unique opportunity for PhD and postdoctoral students in probability theory and mathematical physics interested in these subjects to gain in-depth knowledge from leading researchers.
The deadline for the application for participation is January 27, 2025.
Lecture series by:
- Elena Agliari (Sapienza University of Rome)
- Gérard Ben Arous (CIMS, New York University)
- Aukosh Jagannath (University of Waterloo, Canada)
- Andrea Montanari (Stanford University)
Scientific Organizers:
- Anton Bovier (University of Bonn)
- Véronique Gayrard (CNRS, Aix Marseille University/Bonn Research Chair)
- Giulia Sebastiani (University of Bonn)