Efficient parameter inference in general hidden Markov models using the filter derivatives
2016 (English)In: ICASSP, IEEE International Conference on Acoustics, Speech and Signal Processing - Proceedings, Institute of Electrical and Electronics Engineers (IEEE), 2016, 3984-3988 p.Conference paper (Refereed)
Estimating online the parameters of general state-space hidden Markov models is a topic of importance in many scientific and engineering disciplines. In this paper we present an online parameter estimation algorithm obtained by casting our recently proposed particle-based, rapid incremental smoother (Paris) into the framework of recursive maximum likelihood estimation for general hidden Markov models. Previous such particle implementations suffer from either quadratic complexity in the number of particles or from the well-known degeneracy of the genealogical particle paths. By using the computational efficient and numerically stable Paris algorithm for estimating the needed prediction filter derivatives we obtain a fast algorithm with a computational complexity that grows only linearly with the number of particles. The efficiency and stability of the proposed algorithm are illustrated in a simulation study.
Place, publisher, year, edition, pages
Institute of Electrical and Electronics Engineers (IEEE), 2016. 3984-3988 p.
Hidden Markov models, maximum likelihood estimation, online parameter estimation, particle filters, sequential Monte Carlo methods
IdentifiersURN: urn:nbn:se:kth:diva-195036DOI: 10.1109/ICASSP.2016.7472425ScopusID: 2-s2.0-84973316034ISBN: 978-147999988-0OAI: oai:DiVA.org:kth-195036DiVA: diva2:1044486
41st IEEE International Conference on Acoustics, Speech and Signal Processing, ICASSP 2016, Shanghai International Convention Center Shanghai, China, 20 March 2016 through 25 March 2016
QC 201611032016-11-032016-11-012016-11-03Bibliographically approved