Accelerating chromosome evaluation for partial abductive inference in Bayesian networks by means of explanation set absorption Original Research Article

Partial abductive inference in Bayesian belief networks (BBNs) is intended as the process of generating the K most probable configurations for a set of unobserved variables (the explanation set). This problem is NP-hard and so exact computation is not always possible. In previous works genetic algorithms (GAs) have been used to solve the problem in an approximate way by using exact probabilities propagation as the evaluation function. However, although the translation of a partial abductive inference problem into a (set of) probabilities propagation problem(s) enlarges the class of solvable problems, it is not enough for large networks. In this paper we try to enlarge the class of solvable problems by reducing the size of the graphical structure in which probabilities propagation will be carried out. To achieve this reduction we present a method that yields a (forest of) clique tree(s) from which the variables of the explanation set have been removed, but in which configurations of these variables can be evaluated. Experimental results show a significant speedup of the evaluation function when propagation is performed over the obtained reduced graphical structure.