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:Approximation strategies for structure learning in
Bayesian networks - Teppo NiinimÃ¤ki (Helsinki)
DTSTART;TZID=Europe/London:20160607T093000
DTEND;TZID=Europe/London:20160607T103000
UID:TALK66503AThttp://talks.cam.ac.uk
URL:http://talks.cam.ac.uk/talk/index/66503
DESCRIPTION:Structure discovery in Bayesian networks has attra
cted considerable interest in the recent decades.
Attention has mostly been paid to finding a struct
ure that best fits the data under certain criterio
n. The optimization approach can lead to noisy and
partly arbitrary results due to the uncertainty c
aused by a small amount of data. The so-called ful
l Bayesian approach addresses this shortcoming by
learning the posterior distribution of structures.
In practice\, the posterior distribution is summa
rized by constructing a representative sample of s
tructures\, or by computing marginal posterior pro
babilities of individual arcs or other substructur
es. The state-of-the-art sampling algorithms draw
orderings of variables along a Markov chain. We ha
ve proposed several improvements to these algorith
ms. In this talk I discuss these improvements.
LOCATION:Engineering Department\, CBL Room BE-438
CONTACT:Zoubin Ghahramani
END:VEVENT
END:VCALENDAR