A DBN inference algorithm using junction tree

Dynamic bayesian networks (DBNs) is a compact representation of complex stochastic processes and has been used for many purposes, whose practical application is based on the inference in them. In this paper, we define an optimal node set to d-separate the last slice from the next slice in DBNs, forward interface. Based on that, we present a simple and efficient algorithm - interface algorithm, which implements the forwards and backwards operators using the junction tree algorithm. The interface algorithm uses the junction tree structure constructed from a modified two-slice temporal Bayes net. Finally, we perform complexity analysis for the interface algorithm and give out the lower and upper bounds on the complexity of the interface algorithm.