Learning and Optimization in High Dimensions (StatPhys29 Satellite Workshop)
Via G. Roentgen 1, Milan, Italy

Many foundational challenges in AI, machine learning, inference, and optimization stem from navigating complex, high-dimensional landscapes. Over recent decades, the statistical physics of disordered systems has contributed numerous analytical and numerical frameworks to address these high-dimensional problems, where data enters as a quenched disorder. The relevance of these research directions is underscored by two recent Nobel prizes awarded to Giorgio Parisi and John Hopfield.
Our StatPhys29 Satellite workshop, taking place at Bocconi University on July 9th-11th 2025, will gather leading physicists, mathematicians, computer scientists, and computational neuroscientists to discuss recent advancements and exchange insights on pressing shared open problems.
The talks will be by invitation only, but applicants will have the opportunity to present their own research during a poster session. Attendance will be free for students and researchers, while the Department of Computing Sciences at Bocconi University will provide financial support for the event.
How to reach the Workshop
The workshop will be held in the Grafton Building at Bocconi University, in Via Roentgen 1, Milan. The building is located near Porta Lodovica (Trams 9 & 15) and Porta Romana or Crocetta (Metro 2, yellow line). Buses 59, 61, 79 stop in front of the Building. Circular trams 90, 91 stop 10 minutes walk from campus.
Talks will take place in Room AS-01, which can be reached by going down the stairs directly in front of the GUD Bar, situated below the Grafton Building near the campus entrance. Check the position of the room on Google Maps (remember to go down the stairs).
Poster sessions and coffee breaks will also be held in the same area.
Presenting a Poster at the Workshop
If you are presenting a poster:
The recommended poster size is A1 format (59.4 × 84.1 cm).
You can get your poster printed here in Milan, before your arrival! For this purpose you need to contact MASTERCOPY copy shop at the latest 1 week before the workshop. Please send an email to print@mastercopymilano.it with a clear subject line such as "Poster printing for StatPhys29 workshop at Bocconi - [Name] [Surname]". The shop will provide a price estimate based on the size and type of paper you choose.
Invited Speakers
Maria Chiara Angelini - La Sapienza
Stefan Boettcher - Emory University
Silvio Franz - Università del Salento
Federica Gerace - Università di Bologna
Sebastian Goldt - SISSA
Haiping Huang - Sun Yat-Sen University
Alessandro Ingrosso - Radboud University
Florent Krzakala - EPFL
Clarissa Lauditi - Harvard
Bruno Loureiro - École Normale Supérieure
Marc Mézard - Bocconi
Francesca Mignacco - Princeton
Rémi Monasson - École Normale Supérieure
Mauro Pastore - ICTP
Cengiz Pehlevan - Harvard
Federico Ricci-Tersenghi - La Sapienza
Pietro Rotondo - Università di Parma
Beatriz Seoane - Universidad Complutense de Madrid
Pierfrancesco Urbani - Paris Saclay
Alexander van Meegen - EPFL
Matthieu Wyart - Johns Hopkins
Hajime Yoshino - Osaka University
Schedule
Download the Book of Abstracts of the workshop here !
Wednesday 9th
09:00 - 09:30 REGISTRATION
09:30–10:15 Mézard
Understanding Generative Diffusion: A Statistical-Physics Approach
10:15–11:00 Seoane
Theoretical Insights into Overfitting and Regularization in Energy-Based Models
11:00–11:30 COFFEE BREAK
11:30–12:15 Ricci-Tersenghi
Computing algorithmic thresholds for hard combinatorial problems
12:15–13:00 Urbani
Dynamical decoupling between generalization and overfitting in overparametrized two-layer neural networks
13:00–14:30 LUNCH
14:30–15:00 Huang
Physics of in-context learning in transformers
15:00–15:30 Lauditi
Adaptive kernel predictors from feature-learning infinite limits of neural networks
15:30–16:00 Rotondo
Kernel Renormalization in Bayesian Deep Neural Networks
16:00–17:15 COFFEE BREAK
17:15–18:30 POSTER SESSION 1
18:30–20:00 HAPPY HOUR
Thursday 10th
09:30–10:15 Wyart
Creativity in generative AI
10:15–11:00 Monasson
Energy and computation(s): two examples from natural neural networks
11:00–11:30 COFFEE BREAK
11:30–12:00 Franz
Random features and polynomial rules: a Random Matrix Approach
12:00–12:30 Boettcher
Vectorized implementation of the extremal optimization heuristic
12:30–13:00 Yoshino
Statistical mechanics of supervised learning by multi-layer perceptrons
13:00–14:30 LUNCH
14:30–15:00 Ingrosso
Statistical mechanics of transfer learning in the proportional limit
15:00–15:30 Goldt
Feature learning from non-Gaussian inputs: the case of Independent Component Analysis in high dimensions
15:30–16:00 Angelini
Algorithmic thresholds in Bayesian Inference
16:00–16:30 COFFEE BREAK
16:30–18:00 POSTER SESSION 2
20:00 onward SOCIAL DINNER @ Cascina Cuccagna, Via Cuccagna 2
Friday 11th
09:30–10:15 van Meegen
Symmetries in Neural Loss Landscapes and Their Consequences
10:15–11:00 Krzakala
How and What do neural Neural Networks Learn in High Dimensions?
11:00–11:30 COFFEE BREAK
11:30–12:15 Loureiro
High-dimensional limit of SGD for sequence single-index models
12:15–13:00 Pehlevan
A Solvable Model of In–Context Pretrain–Test Task Alignment
13:00–14:30 LUNCH
14:30–15:00 Mignacco
Optimal protocols for high dimensional learning
15:00–15:30 Gerace
Generalization performance of narrow one-hidden layer networks in the teacher-student setting
15:30–16:00 Pastore
Statistical mechanics of extensive-width Bayesian neural networks near interpolation
16:00–17:15 COFFEE BREAK
Organizers
Carlo Baldassi
Carlo Lucibello
Enrico Malatesta
Gabriele Perugini
Luca Saglietti
Enrico Ventura
Riccardo Zecchina