BELIEF COMBINATION AND PROPAGATION IN A LATTICE-STRUCTURED INFERENCE NETWORK

Belief propagation and belief combination procedures based on Dempster-Shafer's belief function for inference in rule-based system is proposed. The proposed belief combination procedure yields identical results as those of Dempster's rule when evidence are independent. Dempster's rule is shown to be nonrobust when combining evidence with high degree of conflict. The cause of the nonrobustness is discussed and an alternative belief combination procedure is proposed to remedy the deficiency. Although the proposed belief propagation procedure yields results that are dependent upon the interpretations of the rule, the belief propagation procedure is shown to be an interpolation between total ignorance [0,1] and the uncertainty associated with the rule regardless of the interpretations. When the rule interpretation yields associative belief propagation procedure, corresponding chaining syllogism for belief propagation procedure can be derived. Finally, the proposed inference procedures are applied to lattice-structured inference network. © 1990 IEEE