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:Particle filters and curse of dimensionality - Pat
rick Rebeschini (Princeton)
DTSTART;TZID=Europe/London:20140221T140000
DTEND;TZID=Europe/London:20140221T150000
UID:TALK51053AThttp://talks.cam.ac.uk
URL:http://talks.cam.ac.uk/talk/index/51053
DESCRIPTION:A problem that arises in many applications is to c
ompute the conditional distributions of stochastic
models given observed data. While exact computati
ons are rarely possible\, particle filtering algor
ithms have proved to be very useful for approximat
ing such conditional distributions. Unfortunately\
, the approximation error of particle filters grow
s exponentially with dimension\, a phenomenon know
n as curse of dimensionality. This fact has render
ed particle filters of limited use in complex data
assimilation problems that arise\, for example\,
in weather forecasting or multi-target tracking. I
n this talk I will show that it is possible to dev
elop “local” particle filtering algorithms whose a
pproximation error is dimension-free. By exploitin
g conditional decay of correlations properties of
high-dimensional models\, we prove for the simples
t possible algorithm of this type an error bound t
hat is uniform both in time and in the model dimen
sion. (Joint work with R. van Handel)
LOCATION:Engineering Department\, CBL Room BE-438
CONTACT:Zoubin Ghahramani
END:VEVENT
END:VCALENDAR