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:Microsoft Research Cambridge\, public talks
SUMMARY:Convergent and Scalable Algorithms for Expectation
Propagation Approximate Bayesian Inference - Matt
hias Seeger\, EPFL
DTSTART;TZID=Europe/London:20120925T150000
DTEND;TZID=Europe/London:20120925T160000
UID:TALK40067AThttp://talks.cam.ac.uk
URL:http://talks.cam.ac.uk/talk/index/40067
DESCRIPTION:Abstract:\n\nThe expectation propagation (or adapt
ive TAP) relaxation stands out among variational r
elaxations of Bayesian inference\, when it comes t
o generality and accuracy of results. It is widely
used in machine learning today.\nApplied to large
scale continuous variable models for inverse prob
lems in imaging and computer vision\, commonly use
d solvers lack convergence proofs and are too slow
to be useful. In this talk\, we describe a novel
EP algorithm which is both provably convergent and
can be scaled up to large densely connected model
s\, drawing a connection between the double loop a
lgorithm of Opper and Winther (JMLR 2005) and earl
ier work by the author on scalable algorithms for
simpler relaxations. Even for problems of moderate
size (such as Gaussian process classification wit
h a few thousand training points)\, the new algori
thm converges at least an order of magnitude faste
r than the standard (sequential) EP algorithm.\n\n
Partly joint work with Hannes Nickisch.\n\n\nSpeak
er information:\nProfessor Matthias Seeger runs th
e Laboratory for Probabilistic Machine Learning (L
APMAP) at EPFL\, http://lapmal.epfl.ch/
LOCATION:Small lecture theatre\, Microsoft Research Ltd\, 7
J J Thomson Avenue (Off Madingley Road)\, Cambrid
ge
CONTACT:Microsoft Research Cambridge Talks Admins
END:VEVENT
END:VCALENDAR