Computational and Applied Mathematics seminar / Half-way seminar

Abstract:

In Bayesian filtering the objective is to estimate a signal based on noisy observations. If the signal and measurement dynamics are nonlinear, then the problem of finding the conditional filtering density is non-trivial. The classical approaches suffer under the curse of dimensionality and there is a demand for approximate filters that scale well in increasing state dimension. Toward this end, the aim of my research project is to develop scalable filters through the use of deep learning. In this talk I introduce the filtering problem very briefly and discuss a deep splitting method and our contributions to this methodology. This half-way seminar is primarily based on our paper “An energy-based deep splitting method for the nonlinear filtering problem” by K.B., A. Andersson, S. Larsson https://arxiv.org/abs/2203.17153 (accepted for publication in Partial Differ. Equ. Appl. ).