Giulio Biroli
(École Normale Supérieure, Paris)

 
 
Giulio Biroli is professor of theoretical physics at the École Normale Supérieure in Paris.
From 2002 to 2018 he held a researcher position at the Institute for Theoretical Physics (CEA) Giulio Biroli has a long-standing interest in slow dynamics of classical and quantum systems, in disordered and glassy systems and in inter-disciplinary applications of statistical physics ranging from computer science, probability theory, and theoretical ecology. In recent years, part of his research has been devoted to the theory of machine learning leveraging ideas and methods rooted in statistical physics.

Abstract:

Generative AI represents a groundbreaking development within the broader “Machine Learning Revolution,” significantly influencing technology, science, and society. In this talk, I will focus on remarkable connections between generative AI, in particular diffusion models, and statistical physics. Diffusion models are the state of the art to generate images, videos, and sounds. They are very fascinating algorithms for physicists, as they are very much connected to concepts from stochastic thermodynamics, particularly time-reversed Langevin dynamics. These diffusion models start from a simple white noise input and make it evolve through a Langevin process to generate complex outputs such as images, videos, and sounds. I will show that statistical physics provides principles and methods to characterise this generation process. The emergence of features can be understood through the lens of symmetry breaking and dynamical phase transitions, whereas the emergence of creativity - or its absence (memorisation) - can be explained using  theories of disordered systems, and dynamics in complex energy landscapes.

I will conclude by discussing the fundamental reasons behind the success of diffusion models compared to earlier generative methods, highlighting a deep connection between their multi-scale generation mechanism and the renormalisation group analysis of physical systems. 

- - -
 

Refreshments will be served in the lobby of the Leonardo Building after the talk.

- - -
The event will be livetreamed at:
ictp.it/livestream