Research On Bayesian Network Structure Learning

Bayesian network is a graph model which combines probability and statistics with graph theory.It provides powerful help for the expression and reasoning of uncertain knowledge.It has been successfully applied in many fields such as bioinformatics computing,medical treatment,information retrieval and image processing,network security,etc.Among them,Bayesian network learning is the key problem in this field,including structure learning and parameter learning.In this part,because structural learning is the basis of parameter learning,researchers pay more attention to structural learning algorithms.The earliest structural learning is constructed directly by experts relying on domain knowledge.But with the increase of nodes,the relationship between variables becomes more and more complex.It is very difficult or even impossible to construct Bayesian networks only relying on domain knowledge of experts.Therefore,it is impossible to construct Bayesian networks from data.Learning Bayesian network structure has become a key and difficult problem in this field.Because of the large search space,this problem has been proved to be a NPhard problem.Most researchers are concerned about how to find an approximate solution.However,with the deepening of research,many algorithms for finding exact solutions have been proposed in recent years.These algorithms can be used in some places where precise reasoning is needed.At present,although the exact solution algorithm can guarantee the optimal solution,it can not be executed because of the limitation of space and time.So researchers put forward some arbitrary time algorithms.These algorithms transform the original single-round algorithm into multi-round algorithm to achieve a better solution in each round,and run until the time and space permit.To an optimal solution.The work of this paper is as follows:Aiming at the improvement of approximate solution algorithm,after introducing the mathematical model of Phytophthora multicephala and its important value in the field of artificial intelligence in detail,this paper proposes a Bayesian network structure learning algorithm based on Phytophthora multicephala.This algorithm is a hybrid algorithm.Firstly,it aims at the large search space,and according to the fact that Phytophthora multicephala has the weight reserved in the process of network evolution.The original search space is reduced by combining the mathematical model of Phytophthora multicephala and the theory of conditional mutual information.After that,the undirected graph is taken as the basic framework of the network.A better network structure is obtained by using the hill-climbing method based on scoring search to determine the direction of the framework.Finally,in order to get a better network structure,the network is used.The network structure is topologically sorted to get the corresponding order of variables as the input of K2 algorithm to get the final network.Experiments show that the proposed algorithm has shorter running time than the original hill climbing method and the final network score results are similar to the original hill climbing method.Aiming at the improvement of the algorithm for Solving Exact solutions,on the basis of introducing the anytime A* algorithm for solving the optimal Bayesian network in detail,aiming at the shortcomings of the inaccuracy of the previous solution obtained by the algorithm and the large number of generating nodes,this paper proposes to add the sequence information of variables into the heuristic information of anytime A* and use the sequence of variables obtained by the construction algorithm based on polycephalus to deal with anytime.The search direction of me A* algorithm in sequence graph is guided.Experiments show that the algorithm can shorten the upper bound of scoring score of network structure more quickly,so it can reduce the access and generation of poorly scored nodes more,and reduce the memory overhead of the algorithm.