BEGIN:VCALENDAR
VERSION:2.0
PRODID:-//Talks.cam//talks.cam.ac.uk//
X-WR-CALNAME:Talks.cam
BEGIN:VEVENT
SUMMARY:Sampling with diffusion models - Shreyas Padhy\, Jiajun He (Univer
 sity of Cambridge)
DTSTART:20250226T110000Z
DTEND:20250226T123000Z
UID:TALK228910@talks.cam.ac.uk
CONTACT:120952
DESCRIPTION:In this talk\, Shreyas and Jiajun will discuss sampling with d
 iffusion models. We cover two cases\, one is to do posterior (or condition
 ing) sampling of diffusion models\, for applications in inverse imaging\, 
 class-conditional sampling\, text-to-image guidance and finetuning\, and a
 nother is to learn a diffusion models to draw samples from unnormalized de
 nsity.\nFor the former\, we cover inference-only corrections to existing d
 iffusion models that fall under the umbrella of “reconstruction guidance
 ” (DPS\, Red-diff etc)\, as well as training methods such as classifier 
 and classifier-free guidance. Finally\, we discuss some recent work for ef
 ficient finetuning (ControlNet\, DEFT etc)\, as well as an introduction to
  stochastic control techniques (DEFT\, Adjoint Matching).\nFor the latter\
 , we will introduce some recently developed diffusion-based neural sampler
 s\, including diffusion denoising samplers (DDS\, iDEM\, etc) \, escorted 
  samplers (CMCD\, etc.)\, or other variations.
LOCATION:Cambridge University Engineering Department\, CBL Seminar room BE
 4-38.
END:VEVENT
END:VCALENDAR