Matrix-vector algorithms of global posteriori inference in algebraic Bayesian networks

• Published in: 2017 XX IEEE International Conference on Soft Computing and Measurements (SCM) pp. 22 - 24
• Main Authors: Zolotin, Andrey A., Tulupyev, Alexander L.
• Format: Conference Proceeding
• Language: English
• Published: IEEE 01.05.2017
• Subjects: algebraic Bayesian networks; Bayes methods; global inference; ideal of conjuncts; Inference algorithms; Mathematical model; matrix-vector equations; Particle separators; probabilistic graphical models; Probabilistic logic; Software algorithms; Uncertainty
• DOI: 10.1109/SCM.2017.7970483

Algorithm of global posteriori inference in algebraic Bayesian networks is considered in the paper. The results obtained earlier for local a posteriori inference are briefly presented. Main steps of global propagation algorithm are described in details. A transition matrix from the vector of knowledge pattern elements to the virtual evidence, propagated to the next knowledge pattern, is proposed. The stated theorem describes the matrix-vector representation of the stochastic evidence propagation algorithm within a network with scalar estimates of the knowledge patterns elements probabilities of truth. The obtained results form the basis for development of the global posteriori inference machine matrix-vector representation in algebraic Bayesian networks and simplify its further software implementation.