Rao-Blackwellized PSAEM

The Rao-Blackwellized PSAEM is an algorithm for Maximum Likelihood identification of jump Markov linear systems (switching linear systems). The method builds upon recent developments of particle filters.

More precise, the Rao-Blackwellized PSAEM algorithm is an adaption to jump Markov linear models of the general PSAEM algorithm. The PSAEM algorithm was presented in Fredrik Lindsten, An efficient stochastic approximation EM algorithm using conditional particle filters, In Proceedings of the IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP), pp. 6274-6278, Vancouver, Canada, May, 2013.

The main publication concerning the Rao-Blackwellized PSAEM algorithm is

• Andreas Svensson, Thomas B. Schön and Fredrik Lindsten, Identification of jump Markov linear models using particle filters. In Proceedings of the 53rd IEEE Conference on Decision and Control (CDC), Los Angeles, CA, USA, December, 2014. (accepted for publication) PDF, arXiv.org

• Presentation at CDC 2014 (December 17, Los Angeles, CA, USA)

• Poster from ERNSI 2014 (September 24, Oostende, Belgium) (Download full PDF version)

• Presentation at Reglermöte 2014 (June 3, Linköping, Sweden)

Matlab code related to Rao-Blackwellized PSAEM.

RB_PSAEM.zip (.m-code, 11 kb)

The code consists of two parts; run_me is the main script, and RB_PSAEM_JMLGSS, PSAEM_JMLGSS and PSEM_JMLGSS are functions implementing Rao-Blackwellized PSAEM for jump Markov linear systems (hence JMLGSS in the name), PSAEM for jump Markov linear systems and PSEM for jump Markov Linear systems, respectively. These functions were used to produce the numerical examples in the publication.