Ensemble Kalman Filtering
2019-01-24
Talk on Ensemble Kalman Filtering algorithm.
The traditional Kalman filtering algorithm is based on the assumption that all underlying probability densities are Gaussian, which in general is not warranted. On the other hand, a full fledged particle filtering algorithm may require a huge amount of particles to be reliable, which sometimes makes it unfeasible. Ensemble Kalman filtering is an approximate compromise between the Gaussian approximation and pure particle methods, working well with a relatively small particle ensemble, yet not assuming normal distributions. In this talk, an optimization-based justification for the algorithm is given, and its applicability is discussed in the light of computed examples.
