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 Reading Group @ CUED
SUMMARY:LP relaxations for MAP inference - Adrian Weller (
University of Cambridge)
DTSTART;TZID=Europe/London:20151022T143000
DTEND;TZID=Europe/London:20151022T160000
UID:TALK62015AThttp://talks.cam.ac.uk
URL:http://talks.cam.ac.uk/talk/index/62015
DESCRIPTION:For discrete graphical models\, we consider the co
mbinatorial optimization challenge of finding a mo
de configuration of variables\, that is a setting
of all variables that has highest probability\, al
so known as maximum a posteriori (MAP) inference.
We shall provide a brief introduction to a popular
method that frames the problem as an integer line
ar program then relaxes this to a linear program (
LP) over continuous variables. For computational e
fficiency\, the space over which this LP is perfor
med is typically relaxed to an outer bound called
the local polytope which enforces only pairwise co
nsistency. We shall also discuss tighter relaxatio
ns that have recently been explored with some succ
ess\, and touch on message passing methods that ma
y be used to try to solve the problem efficiently.
\n\nreadings:\n\n"Wainwright and Jordan 2008 Graph
ical models\, exponential families and variational
inference Section 8 (p. 195)":https://www.eecs.be
rkeley.edu/~wainwrig/Papers/WaiJor08_FTML.pdf\n\n"
David Sontag's phd thesis chapter 2":http://cs.nyu
.edu/~dsontag/papers/sontag_phd_thesis.pdf
LOCATION:Engineering Department\, CBL Room 438
CONTACT:Yingzhen Li
END:VEVENT
END:VCALENDAR