zbMATH — the first resource for mathematics

Learning technique of probabilistic graphical models: a review. (Chinese. English summary) Zbl 1313.68108
Summary: Probabilistic graphical models are powerful techniques to deal with uncertainty inference efficiently, and learning probabilistic graphical models exactly and efficiently from data is the core problem to be solved for the application of graphical models. Since the representation of graphical models is composed of parameters and structure, their learning algorithms are divided into parameters learning and structure learning. In this paper, the parameters and structure learning algorithms of probabilistic graphical models are reviewed. In parameters learning, the dataset being complete or not is also considered. Structure learning algorithms are categorized into six principal classes according to their different characteristics. The parameters and structure learning of Markov networks are also presented. Finally, the open problems and a discussion of the future trend of probabilistic graphical models are given.

68T05 Learning and adaptive systems in artificial intelligence
68-02 Research exposition (monographs, survey articles) pertaining to computer science
Full Text: DOI