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:Statistics
SUMMARY:MCMC for doubly-intractable distributions - Iain M
urray (University of Toronto)
DTSTART;TZID=Europe/London:20080523T140000
DTEND;TZID=Europe/London:20080523T150000
UID:TALK11789AThttp://talks.cam.ac.uk
URL:http://talks.cam.ac.uk/talk/index/11789
DESCRIPTION:Markov chain Monte Carlo (MCMC) is a well-establis
hed framework for sampling from complex probabilit
y distributions. However\, standard MCMC algorithm
s cannot sample from "doubly-intractable" distribu
tions. \nDoubly-intractable distributions include
the posterior over parameters of many undirected g
raphical models and some point-process models. Eve
ry step of a Markov chain seems to require the com
putation of an intractable normalization term. \n\
nThere are a growing number of valid MCMC algorith
ms for doubly-intractable distributions. They all
involve daunting computations\, but at least give
insight into the problem. I will review what is po
ssible and the implications for the Bayesian learn
ing of undirected graphical models. \n\nIf time al
lows I will share a recent insight by Ryan Adams\,
which combined with MCMC algorithms for doubly-in
tractable distributions\, allows Bayesian density
estimation using Gaussian Processes. \n\nThis is w
ork with David MacKay\, Zoubin Ghahramani and Ryan
Adams. \n\n
LOCATION:MR12\, CMS\, Wilberforce Road\, Cambridge\, CB3 0W
B
CONTACT:
END:VEVENT
END:VCALENDAR