Workshop "Physics of Machine Learning" for Physics of Data

Statistical Artificial Intelligence or Machine Learning is deeply transforming our society and having a relevant impact in several fields including: 1) in the sciences there is growing interest in the predictive ability of models based on the analysis of big data; 2) In the economic sphere, digital transformation (i.e. the process of integrating digital technologies into all aspects of business), is leading to substantial changes at the level of technology and value generation; 3) in social life also promoting customs, habits, and behavioral patterns.

A big issue is that today some of the most accurate models in the Machine Learning world are what we call a ‘black box’ model as they lack explicability and interpretability. As Physicists, this is one of the most exciting challenges and that is why we will devote the second Physics of Data Workshop on this topic, inviting world class experts and young researchers in this field,  providing an overview of the possible open problems and research paths in this field.

Food and accommodation in Asiago will be covered thanks to the UNIPD grant in teaching innovation. Limited number of places available, preferences will be given to first year Physics of Data students. A bus leaving from Padova to Asiago will be organized  and freely available to students and speakers.

Organizers: S. Suweis, M. Baiesi, M. De Domenico, M. Allegra,  and M. Zanetti

Workshop organizational support: Sonia Gelain


22 - 24September 2022


Asiago Astrophysical Observatory, Asiago Italy

A bus for students and speakers is organized and will leave from Largo Egidio Meneghetti, 1, 35131 Padova PD at 10.45 am. 

Please be there 15 minutes before the departure.

A Bus is organized also from Asiago to Padova, after the final lunch of Saturday 24 September.

Registration form

As attendee safety is among our primary concerns, we are carefully monitoring national and local public health guidelines and will respond accordingly. Following Italian and UNIPD rules (at this time, they might change), to attend the workshop you need the Green Pass obtained through vaccination (not by swap).

Registration closed, soon participants will be contacted for the result of the selection procedure.


Thursday 22

14.45 – 15.00 Introduction

15.00 – 16.00 Aurelien Decelle – The Restricted Boltzmann Machine: when the Ising model meets Machine Learning
16.00 – 17.00 Beatriz Seoane Bartolomé – Towards interpretable pattern extraction from dataset with RBMs

17.00 – 17.30 Break/discussion

17.30 – 18.30 Marco Gherardi – Data structure and the combinatorics of expressivity
18.30 – 19.00 Sebastiano Ariosto – Universal mean-field upper bound for the generalization gap of deep neural networks

19.30 Dinner

21.00 Visit of the Observatory

Friday 23

9.30 – 10.30 Patrick Rebeschini – Implicit regularization in statistical learning: An overview and some recent results
10.30 – 11.00 Tomas Vaškevi?ius – Aggregation of statistical estimators

11.00 – 11.30 Break/discussion

11.30 – 12.30 Pietro Rotondo – Statistical mechanics of deep learning beyond the infinite-width limit

13.00 – 15.00 Lunch

15.00 – 16.00 Fabrizio Pittorino – Entropic algorithms and flat minima in the learning landscape of neural networks
16.00 – 16.30 Clarissa Lauditi- Learning through atypical “phase transitions” in overparameterized neural networks
16.30 – 17.00 Giovanni Catania – Thermodynamics of bidirectional associative memories

17.00 – 17.30 Break/discussion

17.30 – 18.00 Marco Letizia – Modern large scale kernel methods for high energy physics
18:00 – 18:30 DIscussion

Saturday 24

9.30 – 10.30 Manuel Saenz – The price of ignorance in low rank matrix estimation
10.30 – 11.00 Jean Barbier – Information theoretic limit of low rank matrix estimation with structured noise
11.00 – 12.00 Giacomo Gradenigo – Algorithmic transitions in simplex learning: replica analysis of the Franz-Parisi potential

12.15 Lunch