Topic: Title: Sequential MCMC with multiple adaptive proposals Authors:  Leopoldo Catania - Aarhus BBS (Denmark) [presenting]
Mauro Bernardi - University of Padova (Italy)
Abstract: The increased amount of large dimensional data calls for new algorithms for sequentially extracting signals from data and estimating model parameters. We propose a new Sequential MCMC algorithm that continuously updates the proposal parameters as the new information arrives. The key innovation of the proposed methodology relies on the mixture of transition kernels of the MCMC algorithm. The validity of the method is tested on large dimensional finite mixture and hidden Markov models as well as on dynamic models with time varying parameters.