BEGIN:VCALENDAR
VERSION:2.0
PRODID:-//Talks.cam//talks.cam.ac.uk//
X-WR-CALNAME:Talks.cam
BEGIN:VEVENT
SUMMARY:Diffusion and Score-based Generative Models - Vincent Dutordoir\, 
 Wenlin Chen\, Tor Fjelde (University of Cambridge)
DTSTART:20220119T110000Z
DTEND:20220119T123000Z
UID:TALK168314@talks.cam.ac.uk
CONTACT:Elre Oldewage
DESCRIPTION:**Abstract**\n\nScore-based generative models have recently sh
 own impressive results in generating synthetic data from complex distribut
 ions — becoming a promising alternative to GANs for sampling photorealis
 tic images. The key idea in these models is to reverse an MCMC chain in wh
 ich a white noise sample is gradually *de*noised to obtain a sample from t
 he target density. During training\, the score functions (gradients of log
  probability density functions) on a large number of noise-perturbed data 
 distributions is learned — hence\, the name of the model. In this readin
 g group\, we cover the basics and advantages of score-based generative mod
 els\, their connections to SDEs and the link with diffusion-based models.\
 n\n**Recommended reading**\n\n- Deep Unsupervised Learning using Nonequili
 brium Thermodynamics. Jascha Sohl-Dickstein\, Eric A. Weiss\, Niru Maheswa
 ranathan\, Surya Ganguli. ICML 2015. [Required]\n\n- Denoising Diffusion P
 robabilistic Models. Jonathan Ho\, Ajay Jain\, Pieter Abbeel. NeurIPS 2020
 . [Optional]\n- Generative Modeling by Estimating Gradients of the Data Di
 stribution. Yang Song\, Stefano Ermon. NeurIPS 2019. [Required]\n\n- Score
 -Based Generative Modeling through Stochastic Differential Equations\, Yan
 g Song\, Jascha Sohl-Dickstein\, Diederik P. Kingma\, Abhishek Kumar\, Ste
 fano Ermon\, Ben Poole. ICLR 2021. [Optional]\n\nOur reading groups are li
 vestreamed via Zoom and recorded for our Youtube channel. The Zoom details
  are distributed via our weekly mailing list.
LOCATION: Cambridge University Engineering Department\, LR12
END:VEVENT
END:VCALENDAR