51�������

Generative AI and Diffusion Models: a statistical physics perspective  

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, highlighting a deep connection between their multi-scale generation mechanism and the renormalisation group analysis of physical systems.