Research On The Bayesian Networks Inference

Bayesian Network based on knowledge representation of probability theory has become a hot spot in the domains of AI uncertain knowledge representation and inference in recent years. Currently many of the domestic and foreign research institutions conducted in-depth study of Bayesian Network. In this paper, the current The research situation of Bayesian Network at home and abroad is analyzed and the principles and construction methods are discussed. The main work and innovations of the paper are as follows:1. The fundamental theory of Bayesian network and conditional independence are introduced. Both the accurate and approximate inference algorithms, such as Message Passing algorithm, Clique Tree Propagation algorithm, Variable Elimination algorithm and so on, are discussed.2. After the discussion of Junction Tree (JT) inference algorithm, an approximate algorithm based on Local Junction Tree (LJT) is proposed for inference of large complex BN. The relevance between a node and the inquiry node Q is defined and is used to construct LJT. The impact of the distance from a node to the node Q and the specialty of the evidence node are considered for the purpose of construction of LJT. In this way, the speed of inference can increase, computing precision can be guaranteed and inference efficiency can be attended at the same time.3. An approximate inference algorithm based on NR is proposed for cropping of DBN. Generally, the process of DBN inference is to update the probability of all nodes. To large complex DBN, the algorithm can get results with good accuracy and increase computing speed evidently.