Rockin’ AI is back!

We are delighted to announce the third edition of Rockin’ AI, following the outstanding success of the first two meetings. Encouraged by the strong scientific engagement of our community, and with the extraordinary support of our guest organizer M. Mondelli, the organizing committee has decided to continue this journey with a new edition.

As Artificial Intelligence becomes increasingly pervasive, driven in particular by the rapid rise of generative models, the need for dedicated discussions on its foundations has never been more pressing. While the technological expansion of AI is unprecedented, many aspects of its functioning remain understood largely at an empirical level.

Despite these open questions, AI systems have achieved remarkable success and are now deeply embedded in everyday technologies. This makes it essential to address fundamental theoretical issues and to develop a principled understanding of the basic building blocks of machine learning models and their training algorithms.

Given the high-dimensional nature of real-world data and the massive number of tunable parameters in modern models, statistical physics and high-dimensional inference provide a natural and powerful framework. Since the seminal works of Gardner and Derrida on the perceptron in the 1980s, the interplay between physics, mathematics, and learning theory has generated deep insights into the behavior of neural networks, a line of research recently recognized by the 2024 Nobel Prize in Physics awarded to G. Hinton and J. Hopfield.

Rockin’ AI is a multidisciplinary conference bringing together researchers from statistical physics, mathematical physics, and machine learning. Over its previous editions, Rockin’ AI has grown into a recurring meeting point for an expanding community at the interface of physics and artificial intelligence. The conference aims to foster cross-disciplinary dialogue and to offer diverse perspectives on key topics in contemporary machine learning, including:

• Associative memories, diffusion models, and energy-based models

• Representation learning and structured data modeling

• Language modeling, self-supervised learning, reasoning, and alternative learning paradigms

• Mathematical physics approaches to high-dimensional probability, spin glasses, and Boltzmann machines

• Theoretical aspects of neural networks

The conference will take place August 31 – Semptember 4  2026, in Roccella Jonica, Calabria (Italy). The event will be hosted at the Convento dei Minimi, offering an inspiring environment for scientific exchange, just steps away from a spectacular Mediterranean beach. In addition to the scientific program, the conference will include a social dinner and group activities during a free afternoon, fostering informal interactions and strengthening community ties.

📅 Applications:  From March 1st to April 30th, 2026, 23:59 (AoE). 

💶 Registration fee: 50EUR (to be paid upon acceptance).

📍 Location: Convento dei Minimi, Largo Molinero, 89047 Roccella Ionica RC.

Confirmed Speakers

Marco Baity-Jesi - ETH Eawag

Adriano Barra - Sapienza University of Rome

Paolo Barucca - University College London

Carlo Baldassi - Bocconi University

Claire Boyer - Université Paris-Saclay

Chiara Cammarota - Sapienza University of Rome

Elisabetta Cornacchia - Bocconi University

Aurelien Decelle - Universidad Politécnica de Madrid

Moritz Helias - Aachen University & Institute for Advanced Simulation

Yue Lu - Harvard University

Marc Mézard - Bocconi University

Theodor Misiakiewicz - Yale University

Rémi Monasson - École Normale Supérieure Paris & CNRS

Andrea Montanari - Stanford University

Valentina Ros - Paris-Saclay & CNRS

Bernd Rosenow - University of Leipzig

Stefano Sarao Mannelli - Chalmers University of Technology

Matthieu Wyart - EPFL Lausanne

Past editions:

ROCKIN' AI - 2024

ROCKIN' AI 2.0 - 2025