Computational advantages of relevance reasoning in Bayesian belief networks

摘要: This paper introduces a computational framework for reasoning in Bayesian belief networks that derives significant advantages from focused inference and relevance reasoning. is based on d-separation other simple computationally efficient techniques pruning irrelevant parts of network. Our main contribution technique we call relevance-based decomposilion, Relevance-based decomposition approaches updating large by focusing their decomposing them into partially overlapping subnetworks. makes some intractable possible and, addition, often results speedup, as the total time taken to update all subnetworks practice considerably less than network whole. We report empirical tests demonstrate practical significance our approach.