BEGIN:VCALENDAR
VERSION:2.0
PRODID:-//talks.cam.ac.uk//v3//EN
BEGIN:VTIMEZONE
TZID:Europe/London
BEGIN:DAYLIGHT
TZOFFSETFROM:+0000
TZOFFSETTO:+0100
TZNAME:BST
DTSTART:19700329T010000
RRULE:FREQ=YEARLY;BYMONTH=3;BYDAY=-1SU
END:DAYLIGHT
BEGIN:STANDARD
TZOFFSETFROM:+0100
TZOFFSETTO:+0000
TZNAME:GMT
DTSTART:19701025T020000
RRULE:FREQ=YEARLY;BYMONTH=10;BYDAY=-1SU
END:STANDARD
END:VTIMEZONE
BEGIN:VEVENT
CATEGORIES:Isaac Newton Institute Seminar Series
SUMMARY:Deep learning as optimal control problems and Riem
annian discrete gradient descent. - Elena Celledon
i (Norwegian University of Science and Technology\
; Norwegian University of Science and Technology)
DTSTART;TZID=Europe/London:20191121T150500
DTEND;TZID=Europe/London:20191121T154500
UID:TALK135052AThttp://talks.cam.ac.uk
URL:http://talks.cam.ac.uk/talk/index/135052
DESCRIPTION:We consider recent work where deep learning neural
networks have been interpreted as discretisations
of an optimal control problem subject to an ordin
ary differential equation constraint. We review th
e first order conditions for optimality\, and the
conditions ensuring optimality after discretisatio
n. This leads to a class of algorithms for solving
the discrete optimal control problem which guaran
tee that the corresponding discrete necessary cond
itions for optimality are fulfilled. The different
ial equation setting lends itself to learning addi
tional parameters such as the time discretisation.
We explore this extension alongside natural const
raints (e.g. time steps lie in a simplex). We comp
are these deep learning algorithms numerically in
terms of induced flow and generalisation ability.
\; References \; - M Ben
ning\, E Celledoni\, MJ Ehrhardt\, B Owren\, CB Sc
hö\;nlieb*\, **Deep learning as optimal control
problems: models and numerical methods**\
,* JCD.

LOCATION:Seminar Room 2\, Newton Institute
CONTACT:INI IT
END:VEVENT
END:VCALENDAR