SciCADE 2013
International Conference on Scientific Computation and Differential Equations
September 16-20, 2013, Valladolid (Spain)

Invited Talk

Show full list of Invited Contributions to Minisymposia Show talk context (MS03)


Can Localization Lift the Curse of Dimensionality for Particle Filters?

S. Reich and Y. Cheng

Abstract
The optimal way of incorporating observations into dynamical models, also known as filtering or data assimilation, can be quite easily done in theory by combining the model dynamics with an application of Bayes' formula. Particle methods come into play as a way to approximate the true filtering solution. The downside is that the number of particles needed to come close to the analytic solution grows exponentially with the model dimension, which is also known as the curse of dimensionality. This problem can in principle be addressed by the concept of localization, which takes advantage of the spatial structure of the state space and transforms a global problem to local problems. This concept has yet to be transferred to particle filters, which we do by using methods from optimal transportation.

Organized by         Universidad de Valladolid     IMUVA