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:The War on Loops - Frederik Eaton (CUED)
DTSTART;TZID=Europe/London:20080911T140000
DTEND;TZID=Europe/London:20080911T153000
UID:TALK13459AThttp://talks.cam.ac.uk
URL:http://talks.cam.ac.uk/talk/index/13459
DESCRIPTION:It is well known that belief propagation is exact
on trees\, i.e. graphs without loops\, although it
often gives accurate results even on graphs with
loops. In this week's RCC I will discuss two appro
aches to improving the accuracy of BP on loopy gra
phs. The relevant papers are:\n\nJM Mooij\, B Wemm
enhove\, HJ Kappen\, T Rizzo\, "Loop Corrected Bel
ief Propagation":http://www.stat.umn.edu/~aistat/p
roceedings/data/papers/042.pdf\n\nM Chertkov\, VY
Chernyak\, "Loop Calculus in Statistical Physics a
nd Information Science":http://arxiv.org/pdf/cond-
mat/0601487\n\nThe first paper describes an algori
thm for propagating cavity distributions which is
exact in graphs with 1 loop. The second paper is a
purely theoretical contribution which gives an ex
pression for the partition function of a factor gr
aph with binary nodes in terms of a finite sum ove
r "generalised loops" in a graphical representatio
n called the "vertex model" (but doesn't describe
an algorithm).\n\nFor further reading\, both paper
s have extended versions:\n\nJ Mooij\, B Kappen\,
"Loop corrections for approximate inference":http:
//arxiv.org/pdf/cs.AI/0612030\n\nM Chertkov\, VY C
hernyak\, "Loop series for discrete statistical mo
dels on graphs":http://arxiv.org/pdf/cond-mat/0603
189\n
LOCATION:Engineering Department\, CBL Room 438
CONTACT:Shakir Mohamed
END:VEVENT
END:VCALENDAR