BEGIN:VCALENDAR
VERSION:2.0
PRODID:-//Talks.cam//talks.cam.ac.uk//
X-WR-CALNAME:Talks.cam
BEGIN:VEVENT
SUMMARY:A tutorial on diffusion models - Emile\, Sasha from CBL
DTSTART:20221109T110000Z
DTEND:20221109T123000Z
UID:TALK192422@talks.cam.ac.uk
CONTACT:86986
DESCRIPTION:Score-based generative models (SGMs) are a powerful class of g
 enerative models that exhibit remarkable empirical performance. Although S
 GMs gained widespread popularity by performing astonishingly well in text-
 to image generation task (DALL-E\, Stable Diffusion)\, it has been shown q
 uite recently that diffusion-based models are able to reach state-of-the-a
 rt quality in many other generative modeling domains in computer vision\, 
 chemistry\, NLP\, and climate modeling. Our talk is designed as a tutorial
  with no prior knowledge on diffusion models required. We will cover i) ba
 sic techniques SGMs rely on (Langevin dynamics and score matching)\; ii) d
 iscrete and then iii) continuous-time diffusion models\; iv) relationship 
 between variational and score-based perspectives.\n\nNo required reading\,
  but we would suggest reading this beforehand for more engagement :) https
 ://arxiv.org/abs/2011.13456\n\nPlease find a non exhaustive list of releva
 nt paper (which we will mention / cover)\n\n• A. Hyvärinen. Estimation 
 of Non-Normalized Statistical Models by Score Matching. 2005\n\n• J. Soh
 l-Dickstein\, E. Weiss\, N. Maheswaranathan\, and S. Ganguli. Deep unsuper
 vised learning using nonequilibrium thermodynamics. 2015\n\n• P. Vincent
 . A connection between score matching and denoising autoencoders. 2011\n\n
 • Y. Song and S. Ermon. Generative modeling by estimating gradients of t
 he data distribution. 2019\n\n• J. Ho\, A. Jain\, and P. Abbeel. Denoisi
 ng diffusion probabilistic models. 2020\n\n• V. De Bortoli\, J. Thornton
 \, J. Heng\, and A. Doucet. Diffusion Schrödinger bridge with application
 s to score-based generative modeling. 2021\n\n• Y. Song\, J. Sohl-Dickst
 ein\, D. P. Kingma\, A. Kumar\, S. Ermon\, and B. Poole. 2021. Score-Based
  Generative Modeling through Stochastic Differential Equations.\n\n• C.-
 W. Huang\, J. H. Lim\, and A. C. Courville. A variational perspective on d
 iffusion-based generative models and score matching. 2021\n\n• Y. Song\,
  C. Durkan\, I. Murray\, and S. Ermon. Maximum likelihood training of scor
 e-based diffusion models. 2021
LOCATION:Cambridge University Engineering Department\, CBL Seminar room BE
 4-38. Zoom link available upon request.
END:VEVENT
END:VCALENDAR