Title: Estimating dynamic networks with a state-space model
Authors: Shaowen Liu - University of Padova (Italy) [presenting]
Massimiliano Caporin - University of Padova (Italy)
Sandra Paterlini - University of Trento (Italy)
Abstract: The aim is to investigate and compare the performance of several types of recursive inference algorithms, including stochastic ensemble Kalman filter (SEKF), ensemble transformation Kalman filter (ETKF) and particle filter (PF), within high-dimensional state-space models, such as the linear Gaussian state-space model and the mixed-Gaussian state-space model. The latter case is especially useful when the true state is a 2D spatial field. In our simulations, we design several field patterns as the hidden state, and examine how the algorithms recover the latent pattern. As for parameter estimation, MCMC is applied. Based on the idea of particle marginal Metroplis-Hastings (PMMH), we propose to use ETKF, instead of particle filter, to generate proposal distribution in each MCMC step. We show that ETKF might be a better choice in case of high-dimensional state-space model.