EVIDENCES PROPAGATIONS IN BAYESIAN NETWORKS

Аннотация

This paper is devoted to some problems of the distribution of several evidences in Bayesian networks.
Currently, there are many different algorithms for calculations in Bayesian networks. Unfortunately, the description
of most algorithms is either absent or only the idea of algorithms is described. Not only algorithms but also ideas for
constructing these algorithms are quite complex. Many questions arise in the process of considering these algorithms
remain unanswered. Some of them can be understood by testing the appropriate software, but many questions remain
unanswered.
We use the idea of dividing the set of network vertices into sets by analogy using the concept of “Generation”.
The concept of "Generation" is convenient to use in the absence of evidence. The presence of evidence requires a
rather complicated adjustment of this concept. However, as a result, the propagation of evidence becomes more
visible, and the corresponding algorithms are greatly simplified.
The presence of several evidences in some cases leads to contradictions, which solutions should be provided for
by the algorithms of the Bayesian network nodes calculations. The modified concept of “Generation” allows one to
find more visual and adequate approaches to resolving contradictions.