Dynamical regimes of diffusion models

Giulio Biroli, Tony Bonnaire (), Valentin Bortoli and Marc Mézard
Additional contact information
Giulio Biroli: ENS, Université PSL, CNRS, Sorbonne Université, Université Paris Cité
Tony Bonnaire: ENS, Université PSL, CNRS, Sorbonne Université, Université Paris Cité
Valentin Bortoli: ENS, CNRS, PSL University
Marc Mézard: Bocconi University

Nature Communications, 2024, vol. 15, issue 1, 1-9

Abstract: Abstract We study generative diffusion models in the regime where both the data dimension and the sample size are large, and the score function is trained optimally. Using statistical physics methods, we identify three distinct dynamical regimes during the generative diffusion process. The generative dynamics, starting from pure noise, first encounters a speciation transition, where the broad structure of the data emerges, akin to symmetry breaking in phase transitions. This is followed by a collapse phase, where the dynamics is attracted to a specific training point through a mechanism similar to condensation in a glass phase. The speciation time can be obtained from a spectral analysis of the data’s correlation matrix, while the collapse time relates to an excess entropy measure, and reveals the existence of a curse of dimensionality for diffusion models. These theoretical findings are supported by analytical solutions for Gaussian mixtures and confirmed by numerical experiments on real datasets.

Date: 2024
References: View complete reference list from CitEc
Citations:

Downloads: (external link)
https://www.nature.com/articles/s41467-024-54281-3 Abstract (text/html)

Related works:
This item may be available elsewhere in EconPapers: Search for items with the same title.

Export reference: BibTeX RIS (EndNote, ProCite, RefMan) HTML/Text

Persistent link: https://EconPapers.repec.org/RePEc:nat:natcom:v:15:y:2024:i:1:d:10.1038_s41467-024-54281-3

Ordering information: This journal article can be ordered from
https://www.nature.com/ncomms/

DOI: 10.1038/s41467-024-54281-3

Access Statistics for this article

Nature Communications is currently edited by Nathalie Le Bot, Enda Bergin and Fiona Gillespie

More articles in Nature Communications from Nature
Bibliographic data for series maintained by Sonal Shukla () and Springer Nature Abstracting and Indexing ().