Efficient Decomposition of Bayesian Networks With Non-graded Variables

Elicitation, estimation and exact inference in Bayesian Networks (BNs) are often difficult because the dimension of each Conditional Probability Table (CPT) grows exponentially with increase number parent variables. The Noisy-MAX decomposition has been proposed to break down a large CPT into several smaller CPTs exploiting assumption causal independence, i.e., absence interaction among In this way, conditional probabilities be elicited or estimated computational burden joint tree algorithm for reduced. Unfortunately, is suited graded variables only, ordinal lowest state as reference, but real-world applications BNs may also involve non-graded variables, like ones reference middle sample space (double-graded variables) two more unordered non-reference states (multi-valued nominal variables). paper, we propose independence decomposition, which includes generalizations double-graded multi-valued While general definition BN implicitly assumes presence all possible interactions, our proposal based on feature that can added upon need. impact investigated published diagnosis acute cardiopulmonary diseases.