BEGIN:VCALENDAR
VERSION:2.0
PRODID:-//Talks.cam//talks.cam.ac.uk//
X-WR-CALNAME:Talks.cam
BEGIN:VEVENT
SUMMARY:SMC Samplers for Applications in High Dimensions - Dr Alexandros B
 eskos\, University College London
DTSTART:20141127T141500Z
DTEND:20141127T151500Z
UID:TALK55633@talks.cam.ac.uk
CONTACT:Fredrik Lindsten
DESCRIPTION:Sequential Monte Carlo (SMC) methods are nowadays routinely ap
 plied in a variety of complex applications: hidden Markov models\, dynamic
 al systems\, target tracking\, control problems\, just to name a few. Wher
 eas SMC methods have been greatly refined in the last decades and are now 
 much better understood\, they are still known to suffer from the curse of 
 dimensionality: algorithms can sometimes break down exponentially fast wit
 h the dimension of the state space. As a consequence\, practitioners in hi
 gh-dimensional Data Assimilation applications in atmospheric sciences\, oc
 eanography and elsewhere will typically use 3D-Var or Kalman-filter-type a
 pproximations that will provide biased estimates in the presence of non-li
 near model dynamics.\n\nThe talk will concentrate on a class of SMC algori
 thms and will look at ways to reduce the cost of the algorithms as a funct
 ion of the dimension of the state space. Explicit asymptotic results will 
 clarify the effect of the dimension at the properties of the algorithm and
  could provide a platform for algorithmic optimisation in high dimensions.
  Applications will be shown in the context of Data Assimilation\, in a pro
 blem where the objective is to target the posterior distribution of the in
 itial condition of the Navier-Stokes equation given a Gaussian prior and n
 oisy observations at different instances and locations of the spatial fiel
 d. The dimension of the signal is in theory infinite-dimensional - in prac
 tice 64x64 or more depending on the resolution – thus posing great chall
 enges for the development and efficiency of SMC methods.
LOCATION:Board Room\, CUED
END:VEVENT
END:VCALENDAR