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:Signal Processing and Communications Lab Seminars
SUMMARY:Adaptive Monte Carlo on multivariate binary sampli
ng spaces - Prof Nicolas Chopin\, ENSAE
DTSTART;TZID=Europe/London:20101012T141500
DTEND;TZID=Europe/London:20101012T151500
UID:TALK26526AThttp://talks.cam.ac.uk
URL:http://talks.cam.ac.uk/talk/index/26526
DESCRIPTION:Authors: Christian SchÃ¤fer (CREST\, CEREMADE)\, Ni
colas Chopin (CREST)\n\nA Monte Carlo algorithm is
said to be adaptive if it can adjust automaticall
y its current proposal distribution\, using past s
imulations. The choice of the parametric family th
at defines the set of proposal distributions is cr
itical for a good performance. We treat the proble
m of constructing such parametric families for ada
ptive sampling on multivariate binary spaces. A pr
actical motivation for\nthis problem is variable s
election in a linear regression context\, where we
need to either find the best model\, with respect
to some criterion\, or to sample from a Bayesian
posterior distribution on the\nmodel space. In ter
ms of adaptive algorithms\, we focus on the Cross-
Entropy (CE) method for optimisation\, and the Seq
uential Monte Carlo (SMC) methods for sampling. Ra
w versions of both SMC and CE\nalgorithms are easi
ly implemented using binary vectors with independe
nt components. However\, for high-dimensional mode
l choice problems\, these straightforward proposal
s do not yields satisfactory\nresults. The key to
advanced adaptive algorithms are binary parametric
families which take at least the linear dependenc
ies between components into account. We review sui
table multivariate binary models\nand make them wo
rk in the context of SMC and CE. Extensive computa
tional studies on real life data with a hundred co
variates seem to prove the necessity of more advan
ced binary families\, to make adaptive Monte Carlo
procedures efficient. Besides\, our numerical res
ults encourage the use of SMC and CE methods as al
ternatives to techniques based on Markov chain exp
loration.\n\nPaper available on arxiv: http://arxi
v.org/abs/1008.0055\n
LOCATION:LR5\, Engineering\, Department of
CONTACT:Rachel Fogg
END:VEVENT
END:VCALENDAR