.png){kind=icon}
.png){kind=icon}
Return
Direct causal structure extraction from pairwise interaction patterns in NAT modeling Bayesian networks
.png){kind=icon}
OA
AI
DOI：10.1016/j.ijar.2018.11.016.png)
.png){kind=icon}
.png){kind=icon}
Share.png){kind=icon}
Save
Abstract
En 中文
Non-impeding noisy-And Trees (NATs) provide a general, expressive, and efficient causal model for conditional probability tables (CPTs) in discrete Bayesian networks (BNs). A CPT may be directly expressed as a NAT model or compressed into a NAT model. Once CPTs are NAT-modeled, efficiency of BN inference (both space and time) can be significantly improved, One of the critical operations in NAT modeling CPTs is extracting NAT structures from interaction patterns between causes. The existing method does so through NAT databases coupled with search trees. Although the databases and search trees are compiled offline, the computation is costly and the dependency of NAT extraction on them adds a resource requirement for online computation. We present a novel method for direct NAT structure extraction from full and valid causal interaction patterns, based on bipartitions of causes. We then extend the method to NAT extraction from partial and invalid interaction patterns. The resultant algorithm suite enables direct NAT extraction from all conceivable practical scenarios, with significantly reduced computational complexity, while eliminating dependency on NAT databases and search trees. (C) 2018 Elsevier Inc. All rights reserved.
Keywords:
Graphical models
Probabilistic inference
Machine learning
Bayesian networks
Causal models
Non-impeding noisy-AND trees
AI Summary
Key information extracted from the uploaded paper, including a brief overview, abstract, background, key highlights, visual analysis, and future outlook.
Journal
IF:
3
Papers:
2.9K
Citations:
5.1K