Abstract: Due to the temporal and spatial complexity in the local inference of multiply sectioned Bayesian networks (MSBN)， an improved algorithm for the local inference of MSBN was proposed. The algorithm redefined the model of MSBN with an object-oriented language. Combined with the concept of vertex degree in graph theory， the algorithm was optimized based on the joint tree algorithm. Considering that the outcome of triangulation was not single， the improved algorithm offered a general solution， which helped to convey message faster and greatly shorten inference time. Finally， an instance of the algorithm was given for experimental analysis， whose results showed that the improved inference algorithm significantly reduces both temporal and spatial complexity.

Key words: multiply sectioned Bayesian network, local inference, joint tree algorithm, vertex degree, triangulation