BEGIN:VCALENDAR
VERSION:2.0
PRODID:www.dresden-science-calendar.de
METHOD:PUBLISH
CALSCALE:GREGORIAN
X-MICROSOFT-CALSCALE:GREGORIAN
X-WR-TIMEZONE:Europe/Berlin
BEGIN:VTIMEZONE
TZID:Europe/Berlin
X-LIC-LOCATION:Europe/Berlin
BEGIN:DAYLIGHT
TZNAME:CEST
TZOFFSETFROM:+0100
TZOFFSETTO:+0200
DTSTART:19810329T030000
RRULE:FREQ=YEARLY;INTERVAL=1;BYMONTH=3;BYDAY=-1SU
END:DAYLIGHT
BEGIN:STANDARD
TZNAME:CET
TZOFFSETFROM:+0200
TZOFFSETTO:+0100
DTSTART:19961027T030000
RRULE:FREQ=YEARLY;INTERVAL=1;BYMONTH=10;BYDAY=-1SU
END:STANDARD
END:VTIMEZONE
BEGIN:VEVENT
UID:DSC-16888
DTSTART;TZID=Europe/Berlin:20200924T110000
SEQUENCE:1600985366
TRANSP:OPAQUE
DTEND;TZID=Europe/Berlin:20200924T120000
URL:https://www.dresden-science-calendar.de/calendar/en/detail/16888
LOCATION:MPI-CBG\, Pfotenhauerstraße 10801307 Dresden
SUMMARY:Hansen: On instabilities\, paradoxes and universal barriers in AI
CLASS:PUBLIC
DESCRIPTION:Speaker: Anders Hansen\nInstitute of Speaker: University of Cam
 bridge\, UK\nTopics:\n\n Location:\n  Name: MPI-CBG (MPI-CBG Auditorium (b
 ig half))\n  Street: Pfotenhauerstraße 108\n  City: 01307 Dresden\n  Phon
 e: +49 351 210-0\n  Fax: +49 351 210-2000\nDescription: Deep learning has 
 had unprecedented success in the sciences\, however\, suffers from a non-n
 egligible Achilles heel: instability. The instability phenomenon in deep l
 earning is universal across different fields ranging from computer vision 
 and image classification\, to voice and audio recognition\, via automated 
 diagnosis in medicine\, to inverse problems and imaging. We will discuss t
 his highly complex issue and provide mathematical explanations for the phe
 nomenon. Intriguingly\, the reasons for the instabilities vary depending o
 n the application. However\, a common phenomenon is that the instabilities
  are not caused by the lack of approximation power of neural networks. Ind
 eed\, it is a paradox that despite the unstable trained networks\, there w
 ill typically exist other stable and accurate networks for the same applic
 ations. The problem is that the training process does not construct them. 
 Herein lies a fascinating barrier. Despite the rich collection of results 
 from approximation theory regarding existence of neural network with power
 ful approximation and stability guaranties\, it can be shown that many of 
 these networks cannot be computed by a computer regardless of computing po
 wer. Thus\, theoretical results a la “there exists a neural network with
  the following properties” do not mean that such a network can ever be c
 omputed on a digital computer.  We are therefore left with the fundamental
  question: can stable and accurate neural networks be computed for the man
 y problems where deep learning is currently used\, or is instability a nec
 essary artefact in AI?
DTSTAMP:20260611T223031Z
CREATED:20200107T230858Z
LAST-MODIFIED:20200924T220926Z
END:VEVENT
END:VCALENDAR