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:Representation\, optimization and generalization p
roperties of deep neural networks - Peter Bartlett
(University of California\, Berkeley)
DTSTART;TZID=Europe/London:20180627T114500
DTEND;TZID=Europe/London:20180627T123000
UID:TALK107434AThttp://talks.cam.ac.uk
URL:http://talks.cam.ac.uk/talk/index/107434
DESCRIPTION:Deep neural networks have improved the state-of-th
e-art performance for prediction problems across a
n impressive range of application areas. This talk
describes some recent results in three directions
. First\, we investigate the impact of depth on re
presentational properties of deep residual network
s\, which compute near-identity maps at each layer
\, showing how their representational power improv
es with depth and that the functional optimization
landscape has the desirable property that station
ary points are optimal. Second\, we study the impl
ications for optimization in deep linear networks\
, showing how the success of a family of gradient
descent algorithms that regularize towards the ide
ntity function depends on a positivity condition o
f the regression function. Third\, we consider how
the performance of deep networks on training data
compares to their predictive accuracy\, we demons
trate deviation bounds that scale with a certain "
spectral complexity\," and we compare the behavior
of these bounds with the observed performance of
these networks in practical problems.

Joint work with Steve Evans\, Dylan Foster\, Dave
Helmbold\, Phil Long\, and Matus Telgarsky.

LOCATION:Seminar Room 1\, Newton Institute
CONTACT:INI IT
END:VEVENT
END:VCALENDAR