ABSTRACT
Dynamic state-space models [24], consisting of a latent Markov process (state/system process) X0, X1, . . . and noisy observations Y1, Y2, . . . that are conditionally independent, are used in a wide variety of applications, for example wireless networks [9], object tracking [21] and econometrics [7], among many others. The model is specified by an initial distribution p(x0|θ), a transition kernel p(xt|xt−1, θ) and an observation distribution p(yt|xt, θ). These distributions are defined in terms of a set of K static (e.g. non time-varying) parameters θ = (θ1, . . . , θK). The joint model to time T is:
p(y1:T ,x0:T , θ) =
( T∏ t=1
p(yt|xt, θ)p(xt|xt−1, θ) ) × p(x0|θ)p(θ), (6.1)
in Bayesian
. . , yT ), etc. These models are also known as hidden Markov models [20].
