For example,Бобцов

ALGEBRAIC BAYESIAN NETWORKS:SEQUESTERED FUSION OF KNOWLEDGE PATTERNS UNDER INFORMATION DEFICIENCY CONDITIONS

Annotation

Subject of Research. The paper deals with situations when two or more models trained on different but crossing datasets are associated with one object in machine learning of probabilistic graphical models. The subject of this study is the fusion of such models, represented by knowledge patterns of algebraic Bayesian network. The study is aimed at description and formalization of the ways for fusion of algebraic Bayesian networks, presented in the form of two knowledge patterns. Method. We created such fusion models that their semantics is clearly explicated by assumptions about the ratio of probabilistic semantics of the considered knowledge patterns. Main Results. We have defined and systematized the ways to fuse knowledge patterns with no generation of new network elements. The statement about the number of atoms in the resulting network and the theorem on the difficulty of maintaining its internal consistency is given and proved. An example of the two networks fusion on a sample with noise is demonstrated. At this, the theoretical distribution of the sample is specified for carrying out comparative analysis, and the sample itself is generated by the Monte Carlo method. Practical Relevance. The methods proposed in the study for algebraic Bayesian networks fusion can be used when applying two or more trained networks describing various properties of a single object. The application of these methods gives the possibility to build a complex network aggregating all accessible data about the object under study and carry out probabilistic-logic operations in it.

Keywords

Постоянный URL

Articles in current issue

CONTENT © ITMO UNIVERSITY 1993-2024