BEGIN:VCALENDAR
VERSION:2.0
PRODID:-//Talks.cam//talks.cam.ac.uk//
X-WR-CALNAME:Talks.cam
BEGIN:VEVENT
SUMMARY:Generative AI and Diffusion Models: a Statistical Physics Analysis
  - Giulio Biroli (CNRS - Ecole Normale Superieure Paris)
DTSTART:20250910T155000Z
DTEND:20250910T163000Z
UID:TALK233302@talks.cam.ac.uk
DESCRIPTION:Generative AI represents a groundbreaking development within t
 he broader &ldquo\;Machine Learning Revolution\,&rdquo\; significantly inf
 luencing technology\, science\, and society. In this talk\, I will focus o
 n the state-of-the-art &ldquo\;diffusion models\,&rdquo\; which are curren
 tly used to generate images\, videos\, and sounds. They are fascinating al
 gorithms 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 t
 hrough a Langevin process to generate complex outputs such as images\, vid
 eos\, and sounds. I will show that statistical physics provides guiding pr
 inciples and methods to characterise this generation process. Specifically
 \, I will discuss how phenomena such as the transition from memorization t
 o generalization and the emergence of data-structure can be understood thr
 ough the lens of symmetry breaking\, phase transitions\, and disordered sy
 stems.&nbsp\;&nbsp\;
LOCATION:External
END:VEVENT
END:VCALENDAR