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:Information-theoretic perspectives on learning alg
orithms - Varun Jog (University of Wisconsin-Madis
on)
DTSTART;TZID=Europe/London:20180222T110000
DTEND;TZID=Europe/London:20180222T120000
UID:TALK101305AThttp://talks.cam.ac.uk
URL:http://talks.cam.ac.uk/talk/index/101305
DESCRIPTION:In statistical learning theory\, generalization er
ror is used to quantify the degree to which a supe
rvised machine learning algorithm may overfit to t
raining data. We overview some recent work [Xu and
Raginsky (2017)] that bounds generalization error
of empirical risk minimization based on the mutua
l information I(S\;W) between the algorithm input
S and the algorithm output W. We leverage these re
sults to derive generalization error bounds for a
broad class of iterative algorithms that are chara
cterized by bounded\, noisy updates with Markovian
structure\, such as stochastic gradient Langevin
dynamics (SGLD). We describe certain shortcomings
of mutual information-based bounds\, and propose a
lternate bounds that employ the Wasserstein metric
from optimal transport theory. We compare the Was
serstein metric-based bounds with the mutual infor
mation-based bounds and show that for a class of d
ata generating distributions\, the former leads to
stronger bounds on the generalization error.

LOCATION:Seminar Room 2\, Newton Institute
CONTACT:info@newton.ac.uk
END:VEVENT
END:VCALENDAR