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 the state-of-the-art “diffusion models,” which are currently used to generate images, videos, and sounds. They are fascinating algorithms for physicists, as they are very much connected to concepts from stochastic thermodynamics, particularly time-reversed Langevin dynamics. Diffusion models initiate from a simple white noise input and evolve it through a Langevin process to generate complex outputs such as images, videos, and sounds. I will show that statistical physics provides guiding principles and methods to characterise this generation process. Specifically, I will discuss how phenomena such as the transition from memorization to generalization and the emergence of data-structure can be understood through the lens of symmetry breaking, phase transitions, and disordered systems.