Bayesian structure learning method based on course learning weight integration

• Main Authors: ZHU CHENG, ZHU XIANQIANG, ZHOU HAOHAO, ZHOU YUN, LIU KAIYUE, LIU LIHUA, HUANG HONGBIN
• Format: Patent
• Language: Chinese; English
• Published: 23.02.2024
• Subjects: CALCULATING; COMPUTER SYSTEMS BASED ON SPECIFIC COMPUTATIONAL MODELS; COMPUTING; COUNTING; PHYSICS

The embodiment of the invention provides a Bayesian structure learning method based on course learning weight integration. The Bayesian structure learning method comprises the steps of sampling a data set to obtain multiple pieces of sampled data; for each piece of sampled data, acquiring an initial course node, and selecting a node, which has the strongest correlation with each node in the course node set, in the candidate node set as a next course node so as to obtain an undirected initial network structure of each piece of sampled data, wherein the candidate node set is a node set except the course node set in the sampled data; performing iterative optimization on the undirected initial network obtained each time through a score optimization function to obtain a weight value of each edge; and integrating the weight values obtained through iterative optimization, and deleting edges of which the weight values are lower than a threshold value so as to obtain an initial directed acyclic graph. The model obtain