A simplified Dynamic Bayesian Network method for modeling repetitive and non-repetitive processes

Dynamic Bayesian networks (DBNs) are increasingly adopted as tools for the modeling of dynamic domains involving uncertainty. Due to their ease of modeling, repetitive DBNs are the standard. However, repetitive DBNs do not allow the independence relations to vary over time. In contrast, non-repetitive DBNs do allow for modeling time-varying relations, but are hard to apply to dynamic domains due to their structural complexity. In this paper, a simplified DBN (sDBN) method is proposed that facilitates the application of nonrepetitive DBNs. The sDBN method achieves this by taking the disjoint repetitive and non-repetitive independence relations as starting points, and by, subsequently, correctly joining these relations. The main benefits of the sDBN method are (i) the qualitative improvement in the graphical representation of DBNs, and (ii) the simplification of the learning procedure of the network. The sDBN method is applied to a real-life task involving time-varying relations and demonstrated to simplify the graphical representation and the learning process. We conclude that the sDBN method facilitates the application of non-repetitive DBNs to dynamic domains.