On the Problem of Performing Exact Partial Abductive Inference in Bayesian Belief Networks using Junction Trees
Luis M. de Campos, José Antonio Gámez, Serafı́n Moral · Studies in fuzziness and soft computing · 2002
Partial abductive inference in Bayesian belief networks has been usually expressed as an extension of total abductive inference (abduction over all the variables in the network). In this paper we study the transformation of the partial problem in a total one, analyzing and trying to improve the method previously appeared in the literature. We also outline an alternative approach, and compare both methods by means of experimentation. The experimental results reveal that the problem of partial abductive inference is difficult to solve by exact computation.