New propagation algorithm in dynamic directed evidential networks with conditional belief functions
Résumé
Proposed as a subclass of directed evidential network with conditional belief functions (DEVN), dynamic directed evidential network with conditional belief functions (DDEVN) was introduced as a new approach for modeling systems evolving in time. Considered as an alternative to dynamic Bayesian network and dynamic possibilistic network, this framework enables to reason under uncertainty expressed in the belief function formalism. In this paper, we propose a new propagation algorithm in DDEVNs based on a new computational structure, namely the mixed binary join tree, which is appropriate for making the exact inference in these networks.