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:Machine Learning @ CUED
SUMMARY:Consensus finding\, exponential models and infinit
e rankings - Dr Marina Meila (University of Washin
gton)
DTSTART;TZID=Europe/London:20081112T130000
DTEND;TZID=Europe/London:20081112T140000
UID:TALK15194AThttp://talks.cam.ac.uk
URL:http://talks.cam.ac.uk/talk/index/15194
DESCRIPTION:This talk is concerned with summarizing -- by mean
s of statistical\nmodels -- of data that expresses
preferences. This data is typically a set of rank
ings of n items by a panel of experts\; the simple
st summary is the "consensus ranking"\, or the "ce
ntroid" of the set of\nrankings. Such problems app
ear in many tasks\, ranging from combining voter p
references to boosting of search engines.\n\nWe st
udy the problem in its more general form of estima
ting a\nparametric model known as the Generalized
Mallows (GM) model. I will present an exact estima
tion algorithm\, non-polynomial in theory\, but ex
tremely effective in comparison with existing algo
rithms. From a statistical point of view\, we show
that the GM model is an exponential family\, and
introduce the conjugate prior for this model class
.\n\nThen we introduce the infinite GM model\, cor
responding to "rankings" over an infinite set of i
tems\, and show that this model is both elegant an
d of practical significance. Finally\, the talk wi
ll touch upon the subject of multimodal distributi
ons and clustering.\n\nJoint work with: Bhushan Ma
ndhani\, Le Bao\, Kapil Phadnis\, Arthur\nPatterso
n and Jeff Bilmes \n\n
LOCATION:Engineering Department\, CBL Room 438
CONTACT:Zoubin Ghahramani
END:VEVENT
END:VCALENDAR