BEGIN:VCALENDAR
VERSION:2.0
PRODID:-//Talks.cam//talks.cam.ac.uk//
X-WR-CALNAME:Talks.cam
BEGIN:VEVENT
SUMMARY:A mathematical theory of deep neural networks - Prof. Helmut Bolcs
 kei\, ETH Zurich
DTSTART:20200522T110000Z
DTEND:20200522T120000Z
UID:TALK134824@talks.cam.ac.uk
CONTACT:Prof. Ramji Venkataramanan
DESCRIPTION:During the past decade deep neural networks have led to specta
 cular successes in a wide range of applications such as image classificati
 on and annotation\, handwritten digit recognition\, speech recognition\, a
 nd game intelligence. In this talk\, we describe efforts to \ndevelop a ma
 thematical theory that can explain these impressive practical achievements
  and possibly guide future deep learning architectures and algorithms. Spe
 cifically\, we develop the fundamental limits of learning in deep neural n
 etworks by characterizing what is possible in principle. We then attempt t
 o explain the inner workings of deep generative networks and of scattering
  networks. A brief survey of recent results on deep networks as solution e
 ngines for PDEs is followed by considerations of interesting open problems
  and philosophical remarks on the role of mathematics in AI research.
LOCATION:Department of Engineering - LT1
END:VEVENT
END:VCALENDAR