On Directed and Undirected Propagation英文文献.pdfVIP

On Directed and Undirected Propagation Algorithms for Bayesian Networks 1 2 3 Christophe Gonzales , Khaled Mellouli , and Olfa Mourali 1 LIP6 – universit´e Paris 6 – France 2 IHEC – Carthage – Tunisie 3 ISIG – Universit´e de Kairouan Abstract. Message-passing inference algorithms for Bayes nets can be broadly divided into two classes: i) clustering algorithms, like Lazy Prop- agation, Jensen’s or Shafer-Shenoy’s schemes, that work on secondary undirected trees; and ii) conditioning methods, like Pearl’s, that use di- rectly Bayes nets. It is commonly thought that algorithms of the former class always outperform those of the latter because Pearl’s-like methods act as particular cases of clustering algorithms. In this paper, a new vari- ant of Pearl’s method based on a secondary directed graph is introduced, and it is shown that the computations performed by Shafer-Shenoy or Lazy propagation can be precisely reproduced by this new variant, thus proving that directed algorithms can be as efficient as undirected ones. 1 Introduction In the last years, Bayesian nets (BN) [1,2,3] have become an increasingly pop- ular knowledge representation framework for reasoning under uncertainty. They combine a directed acyclic graph (DAG) encoding a decomposition of a joint probability distribution over a set of random variables with powerful exact infer- ence techniques [3,4,5,6,7,8,9,10,11,12,13,14]. These can answer various queries including belief updating, i.e., computing the posterior probability of every vari- able given a set of observations, finding the most probable explanation , i.e., find- ing a maxi