EFFICIENT PARTICLE-BASED ONLINE SMOOTHING IN GENERAL HIDDEN MARKOV MODELS

• This paper deals with the problem of estimating expectations of sums of additive functionals under the joint smoothing distribution in general hidden Markov models. Computing such expectations is a key ingredient in any kind of expectation-maximization-based parameter inference in models of this sort. The paper presents a computationally efficient algorithm for online estimation of these expectations in a forward manner. The proposed algorithm has a linear computational complexity in the number of particles and does not require old particles and weights to be stored during the computations. The algorithm avoids completely the well-known particle path degeneracy problem of the standard forward smoother. This makes it highly applicable within the framework of online expectation-maximization methods. The simulations show that the proposed algorithm provides the same precision as existing algorithms at a considerably lower computational cost.

Ämnesord

TEKNIK OCH TEKNOLOGIER  -- Maskinteknik -- Strömningsmekanik och akustik (hsv//swe)

ENGINEERING AND TECHNOLOGY  -- Mechanical Engineering -- Fluid Mechanics and Acoustics (hsv//eng)

Nyckelord

Hidden Markov models

particle filters

smoothing methods

Monte Carlo methods

state estimation

Publikations- och innehållstyp

ref (ämneskategori)

art (ämneskategori)