par Xu, Hòng
Référence Lecture notes in computer science, 945, page (108-116)
Publication Publié, 2014
Article révisé par les pairs
Résumé : Local computational techniques have been proposed to compute marginals for the variables in belief networks or valuation networks, based on the secondary structures called clique trees or Markov trees. However, these techniques only compute the marginal on the subset of variables contained in one node of the secondary structure. This paper presents a method for computing the marginal on the subset that may not be contained in one node. The proposed method allows us to change the structure of the Markov tree without changing any information contained in the nodes, thus avoids the possible repeated computations. Moreover, it can compute the marginal on any subset from the marginal representation already obtained. An efficient implementation of tiffs method is also proposed.