RUS  ENG
Full version
JOURNALS // Vestnik Yuzhno-Ural'skogo Gosudarstvennogo Universiteta. Seriya "Vychislitelnaya Matematika i Informatika" // Archive

Vestn. YuUrGU. Ser. Vych. Matem. Inform., 2023 Volume 12, Issue 1, Pages 61–88 (Mi vyurv293)

Application of tertiary structure of algebraic bayesian network in the problem of a posteriori inference

A. A. Vyatkina, M. V. Abramova, N. A. Kharitonovb, A. L. Tulupyevc

a Saint Petersburg Federal Research Center of the Russian Academy of Sciences (14th line 39, Vasilievsky Island, St. Petersburg, 199178 Russia)
b Saint Petersburg State University (Universitetskaya Emb. 7/9, St. Petersburg, 199034 Russia)
c North-West Institute of Management of the Russian Presidential Academy of National Economy and Public Administration (Sredniy Ave. 57/43, St. Petersburg, 199034 Russia)

Abstract: In the theory of algebraic Bayesian networks, there are algorithms that allow to conduct a global posterior inference using secondary structures. At the same time, building secondary structures implies the use of tertiary structure. Consequently, the question about the separate application of the tertiary structure in the problem of a posterior inference arises. This issue has been considered earlier, but only a general description of the algorithm has been given, and only models with scalar estimates of the probability of truth have been taken into account. In this paper, we present an algorithm that extends the aforementioned algorithm to the possibility of using it in the case of interval estimates. In addition, an important property of an algebraic Bayesian network is acyclicality, and the correctness of the above-mentioned algorithms is ensured only for acyclic networks. Therefore, it is also necessary to be able to check the acyclicity of an algebraic Bayesian network using a tertiary structure. The description of this algorithm is also presented in this paper, it is based on the previously proved theorem that relates the number of knowledge pattern models in the network to the number of non-empty separators and the number of strong restriction connectivity components in acyclic algebraic Bayesian network, as well as the theorem proved in this paper that two knowledge pattern models belong to the same strong restriction connectivity component. For all the developed algorithms, the correctness of their performance is proved, and their time complexity estimation is calculated.

Keywords: algebraic Bayesian networks, knowledge pattern, logical and probabilistic inference, tertiary structure, probabilistic graphical models, machine learning.

UDC: 004.8

Received: 09.12.2022

DOI: 10.14529/cmse230104



© Steklov Math. Inst. of RAS, 2024