Accelerated Quantification of Bayesian Networks with Incomplete Data

Probabilistic expert systemsbased on Bayesian networks(BNs)require initial specification both a qualitative graphical structure and quantitative assessmentof conditional probability tables. Thispaperconsidersstatistical batchlearning of the probability tables on the basis of incomplete data and expert knowledge. The EM algorithm with a generalized conjugate gradient acceleration methodhas been dedicated to quantification of BNsby maximumposterior likelihoodestimationfor a super-class of the recursive graphical models. This newclass of modelsallowsa great varietyof local functionalrestrictions to be imposedon the statistical model, which hereby extents the control and applicability of the constructed methodfor quantifying BNs.

• Type: Conference
• Year: 1995
• Where: KDD
• Authors: Bo Thiesson
• Keywords: Conditional Probability Tables, Data Mining, KDD 1995, Probability Tables, Qualitative Graphical Structure