Open Conference Systems, 50th Scientific meeting of the Italian Statistical Society

Font Size: 
Bayesian inference for hidden Markov models via duality and approximate filtering distributions
Guillaume KON KAM KING, Omiros Papaspiliopoulos, Matteo Ruggiero

Last modified: 2018-05-21


Filtering hidden Markov models is analytically tractable only for a handful of models (e.g. Baum-Welch and Kalman filters).Recently, Papaspiliopoulos \& Ruggiero (2014) proposed another analytical approach exploiting a duality relation between the hidden process and an auxiliary process, called dual and related to the time reversal of the former. With this approach, the filtering distributions are obtained as a recursive series of finite mixtures.Here, we study the computational effort required to implement this strategy in the case of two hidden Markov models, the Cox-Ingersoll-Ross process and the $K$-dimensional Wright-Fisher process, and examine several natural and very efficient approximation strategies.


  1. Cappé, O., Moulines, E., Ryden, T.: Inference in Hidden Markov Model. Springer, New York, USA (2005)
  2. Papaspiliopoulos, O., Ruggiero, M.:Optimal filtering and the dual process.Bernoulli 20, 1999--2019 (2014)
  3. Fieker, C., Hart, W., Hofmann, T., Johansson, F.: Nemo/Hecke: Computer Algebra and Number Theory Packages for the Julia Programming Language. In: Proc. 2017 ACM Int. Symp. Symb. Algebr. Comput., pp. 157--164. ACM, New York, USA (2017)

Full Text: PDF