University of Cambridge > Talks.cam > Signal Processing and Communications Lab Seminars > Filtering and Smoothing in Non-linear Dynamical Systems using Quadrature Expectation Propagation (EP)

Filtering and Smoothing in Non-linear Dynamical Systems using Quadrature Expectation Propagation (EP)

Add to your list(s) Download to your calendar using vCal

If you have a question about this talk, please contact Taylan Cemgil.

The unscented Kalman filter (originally developed at the Cambridge signal processing lab) is a fast approximate filter for non-linear dynamical systems. It is reasonably accurate if the dynamics and observation model are nearly linear. For some models however, we can show that the unscented Kalman filter provably breaks down. To be more precise, for models where the observation model is such that state and observation are uncorrelated (but still dependent) the unscented Kalman filter does not update the state estimate at all after a new observation. We propose quadrature EP, a very general approximate inference technique that is based on expectation propagation and Gaussian quadrature. The special case of filtering in non-linear dynamical systems can be referred to as a one-step unscented 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.

This talk is part of the Signal Processing and Communications Lab Seminars series.

Tell a friend about this talk:

This talk is included in these lists:

Note that ex-directory lists are not shown.

 

© 2006-2020 Talks.cam, University of Cambridge. Contact Us | Help and Documentation | Privacy and Publicity