BEGIN:VCALENDAR
VERSION:2.0
PRODID:-//Talks.cam//talks.cam.ac.uk//
X-WR-CALNAME:Talks.cam
BEGIN:VEVENT
SUMMARY:Divide-and-conquer posterior sampling for Denoising Diffusion prio
 rs - Yazid Janati (École Polytechnique)
DTSTART:20240718T100000Z
DTEND:20240718T103000Z
UID:TALK219019@talks.cam.ac.uk
DESCRIPTION:Recent advancements in solving Bayesian inverse problems have 
 spotlighted denoising diffusion models (DDMs) as effective priors. Althoug
 h these have great potential\, DDM priors yield complex posterior distribu
 tions that are challenging to sample from. Existing approaches to posterio
 r sampling in this context address this problem either by retraining model
 -specific components\, leading to stiff and cumbersome methods\, or by int
 roducing approximations with uncontrolled errors that affect the accuracy 
 of the produced samples. We present an innovative framework\, divide-and-c
 onquer posterior sampling\, which leverages the inherent structure of DDMs
  to construct a sequence of intermediate posteriors that guide the produce
 d samples to the target posterior. Our method significantly reduces the ap
 proximation error associated with current techniques without the need for 
 retraining. We demonstrate the versatility and effectiveness of our approa
 ch for a wide range of Bayesian inverse problems.
LOCATION:External
END:VEVENT
END:VCALENDAR