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:Candidates vs. Noises Estimation for Large Multi-C
lass Classification Problem - Tong Zhang (Rutgers\
, The State University of New Jersey)
DTSTART;TZID=Europe/London:20180628T090000
DTEND;TZID=Europe/London:20180628T094500
UID:TALK107479AThttp://talks.cam.ac.uk
URL:http://talks.cam.ac.uk/talk/index/107479
DESCRIPTION:In practice\, there has been sigificant interest i
n multi-class classification problems where the nu
mber of classes is large. Computationally such app
lications require statistical methods with run tim
e sublinear in the number of classes. A number of
methods such as Noise-Contrastive Estimation (NCE)
and variations have been proposed in recent years
to address this problem. However\, the existing m
ethods are not statistically efficient compared to
multi-class logistic regression\, which is the ma
ximum likelihood estimate. In this talk\, I will d
escribe a new method called Candidate v.s. Noises
Estimation (CANE) that selects a small subset of c
andidate classes and samples the remaining classes
. We show that CANE is always consistent and compu
tationally efficient. Moreover\, the resulting est
imator has low statistical variance approaching th
at of the maximum likelihood estimator\, when the
observed label belongs to the selected candidates
with high probability. Extensive experimental resu
lts show that CANE achieves better prediction accu
racy over a number of the state-of-the-art tree c
lassifiers\, while it gains significant speedup co
mpared to standard multi-class logistic regression
.
LOCATION:Seminar Room 1\, Newton Institute
CONTACT:INI IT
END:VEVENT
END:VCALENDAR