Sangeetika Ruchi (Centrum Wiskunde & Informatica, Amsterdam)
Friday, January 18, 2019 - 10:15
SFB 1294 Seminar, University of Potsdam
Karl-Liebknecht- Str. 24-25, 14476 Potsdam OT Golm, House 9, Room 0.13
Sequential Monte Carlo methods (SMC) are typically stochastic. Ensemble Transform Particle filter (ETPF) is a deterministic SMC method. It, however, still fails for strongly nonlinear problems, since the prior measure does not approximate well the posterior measure leading to the method degeneracy. In this work we propose to mutate ETPF based on a Markov kernel with an invariant measure and to introduce intermediate densities to ETPF to avoid the filter collapse. We show that the adjusted ETPF outperforms a tempering SMC method and a regularized ensemble Kalman filter for non-Gaussian high-dimensional problems of parameter estimation. Invited by Sebastian Reich.
submitted by Liv Heinecke (liv.heinecke@uni-potsdam.de, 0331-977-203137)