Exact inference for a class of hidden Markov models on general state spaces (Q2044399)

The authors derive sufficient conditions for existence of possibility of exact inference (i.e. evaluation of all distributions of interest: filtering, prediction, smoothing and likelihood with a finite computational effort) for a special class of hidden Markov models based on a set of discrete indirect observations linked the signal. These conditions are concerned with the existence of a certain type of dual process which allows representation of the distributions and functions of interest as finite mixtures of elementary densities. The recursive procedures are proposed which allow to estimate the parameters involved, yielding similar results to those obtained with Baum-Welch filters on finite state spaces (see [\textit{O. Cappé} et al., Inference in hidden Markov models. New York, NY: Springer (2005; Zbl 1080.62065)]). A set of practical algorithms are developed for implementing the recursions. The superior performance of the algorithms is illustrated with the particle filters both in accuracy and computational efficiency. The code for optimal filtering, smoothing and parameter inference is made available in the Julia package DualOptimalFiltering (see \url{https://github.com/konkam/DualOptimalFiltering.jl}).