Date of Award


Publication Type

Master Thesis

Degree Name



Computer Science




Bayesian networks are widely used for knowledge representation and uncertain reasoning. One of the most important services which Bayesian networks provide is (probabilistic) inference. Effective inference algorithms have been developed for probabilistic inference in Bayesian networks for many years. However, the effectiveness of the inference algorithms depends on the sizes of Bayesian networks. As the sizes of Bayesian networks become larger and larger in real applications, the inference algorithms become less effective and sometimes are even unable to carry out inference. In this thesis, a new inference algorithm specifically designed for large and complex Bayesian networks, called 'path propagation', is proposed. Path propagation takes full advantage of one of the most popular inference algorithms, i.e., global propagation. It improves over global propagation by carrying out inference only in certain paths in a junction tree that are relevant to queries. Compared with global propagation, path propagationtakes less computational resources and can effectively improve the computational efficiency for inference in large and complex Bayesian networks.