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:Inference Group
SUMMARY:MCMC for doubly-intractable distributions - Iain M
urray\, Gatsby Computational Neuroscience Unit\, U
CL
DTSTART;TZID=Europe/London:20060906T140000
DTEND;TZID=Europe/London:20060906T150000
UID:TALK5134AThttp://talks.cam.ac.uk
URL:http://talks.cam.ac.uk/talk/index/5134
DESCRIPTION:"The model is intractable so we resort to Markov c
hain Monte Carlo" has become a standard mantra in
the Bayesian statistics community. But (even for t
he patient) standard MCMC techniques are not a pan
acea — for example they can not sample from the pa
rameter-posterior of a large tree-width undirected
graphical model.\n\nWe recently (Proc. UAI 2006)
introduced a valid MCMC scheme for this problem. O
ur _exchange algorithm_ is simpler and often perfo
rms better than the only direct competitor (Møller
et al.\, Biometrika 93(2):451–458\, 2006). Althou
gh both require expensive exact sampling (Propp an
d Wilson\, Rand. Struct. Alg. 9(1&2):223–252 1996)
.\n\nIn this talk I give a simpler derivation of t
he exchange algorithm. I also discuss the extent t
o which exact sampling is required and the implica
tions for probabilistic modeling with undirected g
raphs.\n\nThis is work with David MacKay and Zoubi
n Ghahramani.
LOCATION:Ryle Seminar Room\, Cavendish Laboratory
CONTACT:Ryan Prescott Adams
END:VEVENT
END:VCALENDAR