BEGIN:VCALENDAR
VERSION:2.0
PRODID:-//Talks.cam//talks.cam.ac.uk//
X-WR-CALNAME:Talks.cam
BEGIN:VEVENT
SUMMARY:Filtering and Smoothing in Non-linear Dynamical Systems using Quad
 rature Expectation Propagation (EP) - Onno Zoeter\, Microsoft Research Cam
 bridge
DTSTART:20070821T120000Z
DTEND:20070821T130000Z
UID:TALK7703@talks.cam.ac.uk
CONTACT:Taylan Cemgil
DESCRIPTION:The unscented Kalman filter (originally developed at the Cambr
 idge signal processing lab) is a fast approximate filter for non-linear dy
 namical systems. It is reasonably accurate if the dynamics and observation
  model are nearly linear. For some models however\, we can show that the u
 nscented Kalman filter provably breaks down. To be more precise\, for mode
 ls where the observation model is such that state and observation are unco
 rrelated (but still dependent) the unscented Kalman filter does not update
  the state estimate at all after a new observation. \nWe propose quadratur
 e EP\, a very general approximate inference technique that is based on exp
 ectation propagation and Gaussian quadrature. The special case of filterin
 g in non-linear dynamical systems can be referred to as a one-step unscent
 ed Kalman filter. It is just as fast as the unscented Kalman filter\, yet 
 significantly more accurate\, in particular in the problematic model class
  described above.\n
LOCATION:LR6\, Engineering\, Department of
END:VEVENT
END:VCALENDAR