Approximate Inference and Forecast Algorithms in Graphical Models for Partially Observed Dynamic Systems
Proceedings of the Sixth International Workshop on Artificial Intelligence and Statistics, PMLR R1:319-319, 1997.
From a statistical point of view , modelling stochastic temporal processes by graphical models is a suitable choice, specially when certain standard assumptions in classical modelling cannot be assumed. Focusing the discussion on partially observed domains, it is important to design algorithms which provide probability distributions over the current and future states of the non-observable components of the domain, using the information stored in the observable components. In this paper, we present a simulation algorithm for approximating the exact probability distributions associated with such inference and forecast processes. This algorithm uses both the probabilities built at the previous time step and the new evidence obtained to propose new probability distributions associated with current and future states of the domain. To validate the algorithm, a case study of equipment maintenance is considered.