| Abstract
|
A Bayesian belief network uses a directed … A Bayesian belief network uses a directed acyclic graph to represent probabilistic dependencies among a set of variables. Typically, performing inference on a belief network involves computing conditional probabilities among variables in the network. We introduce an algorithm for belief-network inference that is based on recursive decomposition. The algorithm recursively bisects a belief network to create a binary tree. The tree then is used for probabilistic inference. We describe the recursive-decomposition inference algorithm in sufficient detail for it to be implemented readily, and we prove the validity of the algorithm. We also link belief-network inference that is based on recursive decomposition to the literature on vertex separators. The recursive divide-and-conquer nature of the recursive-decomposition inference algorithm allows an implementation that is brief and simple; it also facilitates our analysis of the time complexity of the algorithm. A companion paper contains an analysis and evaluation of the inference algorithm on numerous belief network structures. thm on numerous belief network structures.
|
| Address
|
Stanford, CA, USA +
|
| Author
|
Gregory F. Cooper +
|
| Bibtype
|
techreport +
|
| Institution
|
Knowledge Systems, AI Laboratory +
|
| Key
|
KSL-90-05 +
|
| Modification dateThis property is a special property in this wiki.
|
1 May 2009 13:36:42 +
|
| Month
|
July +
|
| Number
|
KSL-90-05 +
|
| Tag
|
Computer science +
|
| Title
|
Bayesian Belief-Network Inference Using Recursive Decomposition +
|
| Tr id
|
KSL-90-05 +
|
| Year
|
1992 +
|
| Categories |
Technical Report,
Publication,
KSL Technical Report
|
