BEGIN:VCALENDAR
VERSION:2.0
PRODID:-//CERN//INDICO//EN
BEGIN:VEVENT
SUMMARY:Deep Denoising: An Overview
DTSTART:20261029T103000Z
DTEND:20261029T120000Z
DTSTAMP:20261101T120800Z
UID:indico-event-19470@indico.psi.ch
CONTACT:benjamin.bejar@psi.ch
DESCRIPTION:Speakers: Luis Felipe Barba Flores (PSI - Paul Scherrer Instit
 ut)\n\nAbstract: Recovering clean signals from noisy observations is a fou
 ndational problem in scientific imaging\, and deep neural networks have su
 bstantially redefined what is achievable. This talk surveys the landscape 
 of deep denoising methods\, organized around a central question: what supe
 rvision is available at training time? We begin with the fully supervised 
 setting\, where paired clean targets exist\, and trace the progression tow
 ard regimes where clean data is scarce or entirely absent\, covering metho
 ds that exploit pairs of independent noisy realizations\, techniques that 
 learn from a single corrupted image by leveraging statistical independence
  assumptions across pixels or measurements\, and approaches grounded in sc
 ore estimation and Tweedie's formula that connect denoising to the underly
 ing data distribution. We also examine how the structure of the measuremen
 t process itself\, particularly in tomographic and other linear inverse pr
 oblems\, can be turned into a self-supervision signal\, removing the need 
 for any clean reference. We might also discuss diffusion-based posterior s
 amplers\, which shift the goal from a single point estimate to characteriz
 ing the full posterior over clean images given noisy observations. Through
 out\, we highlight the tradeoffs between data requirements\, noise model a
 ssumptions\, and the nature of the outputs produced. Finally\, I will touc
 h upon some applications in computed tomography and how this denoising tec
 hniques enable better data compression. \nfor online participation: TEAMS
  link here \n \n\nhttps://indico.psi.ch/event/19470/
LOCATION:OFLG/402 or online
URL:https://indico.psi.ch/event/19470/
END:VEVENT
END:VCALENDAR