Dynamic Possibilistic Networks: Representation and Exact Inference

In this paper, we present dynamic possibilistic networks (DPNs); a new framework for modeling uncertain sequential data with possibilistic graphical models. Depending on the kind of conditioning, two types of networks are studied here: the product-based and the min-based networks. Hence two versions of an exact algorithm for inference in such networks are described here. The main contribution of this paper is the use of possibility theory as a framework for representation of temporal networks which gives an alternative framework for dynamic probabilistic networks. We especially, present how junction trees can be used to make online inference namely filtering problem in product-based DPNs and min- based DPNs and we will discuss how this technique can be extended to make prediction.