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:On the Bethe approximation - Adrian Weller (Columb
ia University)
DTSTART;TZID=Europe/London:20140811T110000
DTEND;TZID=Europe/London:20140811T120000
UID:TALK53562AThttp://talks.cam.ac.uk
URL:http://talks.cam.ac.uk/talk/index/53562
DESCRIPTION:Belief propagation is a remarkably effective tool
for inference in graphical models\, even when appl
ied to networks with cycles. A variational perspec
tive shows that it may be viewed as a way to seek
the minimum of the Bethe free energy\, though it m
ay converge only to a local optimum or may not con
verge at all.\n\nWe shall cover a brief introducti
on to these ideas\, then go on to describe a recen
t algorithm we developed for any binary pairwise m
odel which\, to our knowledge\, is the first to gu
arantee to return an epsilon-approximation to the
_global minimum_ of the Bethe free energy. The app
roach involves discretizing to yield a discrete op
timization problem\, which may be viewed as multi-
label MAP inference. If the initial model is fully
attractive\, this yields a fully polynomial-time
approximation scheme (FPTAS).\n\nIf time\, we can
also discuss work that further explores the Bethe
approximation and tries to tease apart the two way
s it differs from exact inference: (i) the true en
tropy is approximated by the Bethe (pairwise) entr
opy\, and (ii) the minimization is performed over
a relaxation of the marginal polytope (which enfor
ces a globally consistent probability distribution
) termed the local polytope (which enforces only p
airwise consistency). \n\nThis is joint work with
Tony Jebara at Columbia University.\n\nRelated pap
ers:\nA. Weller and T. Jebara\, "Approximating the
Bethe Partition Function" . Uncertainty in Artifi
cial Intelligence (UAI)\, 2014.\nA. Weller\, K. Ta
ng\, D. Sontag and T. Jebara\, "Understanding the
Bethe Approximation: When and How can it go Wrong?
" . Uncertainty in Artificial Intelligence (UAI)\,
2014.\n
LOCATION:Engineering Department\, CBL Room BE-438.
CONTACT:Zoubin Ghahramani
END:VEVENT
END:VCALENDAR