BEGIN:VCALENDAR
VERSION:2.0
PRODID:-//Talks.cam//talks.cam.ac.uk//
X-WR-CALNAME:Talks.cam
BEGIN:VEVENT
SUMMARY:Universal Adversarial Perturbations: Fooling Deep Networks with a 
 Single Image - Alhussein Fawzi\; UCLA\, DeepMind
DTSTART:20180130T140000Z
DTEND:20180130T150000Z
UID:TALK99730@talks.cam.ac.uk
CONTACT:Frank Kelly
DESCRIPTION:The robustness of classifiers to small perturbations of the da
 ta points is a highly desirable property when the classifier is deployed i
 n real and possibly hostile environments. Despite achieving excellent perf
 ormance on recent visual benchmarks\, I will show in this talk that state-
 of-the-art deep neural networks are highly vulnerable to universal\, image
 -agnostic\, perturbations. After demonstrating how such universal perturba
 tions can be constructed\, I will analyse the implications of this vulnera
 bility and provide a geometric explanation for the existence of such pertu
 rbations via an analysis of the curvature of the decision boundaries.
LOCATION:Centre for Mathematical Sciences\, MR4
END:VEVENT
END:VCALENDAR