ABSTRACT
General state-space models, including non-linear and non-Gaussian dynamic models, are described here. Posterior inference via MCMC algorithms that work well for these types of general models is challenging. Furthermore, MCMC schemes are typically not feasible for real-time (online) filtering and parameter learning in time series settings, since a new target posterior distribution - and therefore a new MCMC run - needs to be considered each time a new observation arrives. This chapter describes and illustrates sequential Monte Carlo approaches as an alternative and computationally feasible tool for online filtering and sequential posterior learning within the class of general state-space models.
