Adaptive dynamic probabilistic networks for distributed uncertainty processing

Uncertainty processing is a core task in many applications of distributed systems. Typical distributed systems have local processing nodes to collect information, which usually contain uncertainties, and do computational work. The nodes can interact with each other, and they evolve with time. A promising way of modelling and processing uncertainties in these systems is to use graphical models to form beliefs about the required information. Dynamic probabilistic networks for distributed uncertainty processing are presented in this paper. Two approaches are given, and comparison shows that the model with the more state- of- art approach performs better. Since it is not possible to obtain enough knowledge to construct an exact model at the beginning, the model needs to adjust itself when evolving. Therefore we have developed a parameter update algorithm to make the model adapt to the changing environment. Experiments are presented to show the effectiveness of the models and the algorithms.