BEGIN:VCALENDAR
VERSION:2.0
PRODID:-//Talks.cam//talks.cam.ac.uk//
X-WR-CALNAME:Talks.cam
BEGIN:VEVENT
SUMMARY:Robustness and Accuracy of Deep End-to-End Networks for Inverse Pr
 oblems - Martin Genzel (Helmholtz-Zentrum Berlin for Materials and Energy)
DTSTART:20220623T140000Z
DTEND:20220623T153000Z
UID:TALK174947@talks.cam.ac.uk
CONTACT:Yury Korolev
DESCRIPTION:In the past five years\, deep learning methods have become sta
 te-of-the-art in solving various inverse problems. Before such approaches 
 can find application in safety-critical fields\, a verification of their r
 eliability appears mandatory. For example\, recent works have pointed out 
 instabilities of deep neural networks for several image reconstruction tas
 ks. In analogy to adversarial attacks in classification\, it was shown tha
 t slight distortions in the input domain may cause severe artifacts. In th
 is talk\, we will shed new light on this concern and deal with a quantitat
 ive robustness analysis of deep-learning-based algorithms for solving unde
 rdetermined inverse problems. This covers compressed sensing with Gaussian
  measurements as well as image recovery from Fourier and Radon measurement
 s\, including a real-world scenario for magnetic resonance imaging (using 
 the NYU-fastMRI dataset). Our main focus is on computing adversarial pertu
 rbations of the measurements that maximize the reconstruction error. Our e
 mpirical results reveal that standard end-to-end network architectures are
  not only surprisingly resilient against statistical noise\, but also agai
 nst adversarial perturbations. Remarkably\, all considered networks are tr
 ained by common deep learning techniques\, without adversarial defense str
 ategies. We will also relate our results to the aspect of accuracy\, which
  is discussed in the context of the 2021 AAPM Sparse-View CT Challenge.\nT
 his is joint work with Ingo Gühring (TU Berlin)\, Maximilian März (Amazo
 n)\, and Jan Macdonald (TU Berlin).
LOCATION:MR14\, Centre for Mathematical Sciences &amp\; Zoom (contact orga
 niser for the zoom link)
END:VEVENT
END:VCALENDAR