Algorithms for high-dimensional non-linear filtering and smoothing problems
In many applicational areas there is a need to improve real-time predictions of the state of a dynamical system of interest via partial observations. Such estimation problems are typically approached with so called filtering algorithms.
Filtering is particularly challenging in the context of non-linear evolution models with an extended state space. Smoothing can be viewed as a form of filtering with an augmented state space which makes the problem even higher dimensional. Here we will present a series of tools for a class of different smoothers algorithms and explore techniques such as localisation (in time and space), hybrid formulations combining complementary smoothers and adaptive spread corrections that can be used to address the difficulties that arise in a high dimensional non-linear problem setting.
Dr. Jana de Wiljes Uni Potsdam