Statistics and Its Interface

Volume 6 (2013)

Number 1

The tree structure of graphs for various graphical models

Pages: 151 – 164

DOI: http://dx.doi.org/10.4310/SII.2013.v6.n1.a14

Authors

Jianhua Guo (School of Mathematics and Statistics, Northeast Normal University, Changchun, China)

Xiaofei Wang (School of Mathematics and Statistics, Northeast Normal University, Changchun, China)

Abstract

After proper decompositions or separations, there is a common characteristic of the secondary structures for various graphical models. In this paper, we show that the junction tree captures this common characteristic. To generalize all potential occurrences in different graphical models, we define junction trees on general set classes and show several equivalent properties of junction trees. For mixed graphical models and hierarchical models, we investigate in detail the M-decomposition of marked graphs and the H-decomposition of interaction graphs, and point out the junction tree structures of marked graphs and interaction graphs. Moreover, properties of separation trees and dseparation trees are discussed for undirected and directed graphs, respectively. Both separation and d-separation trees are closely associated with junction trees. Finally, we propose two algorithms for constructing junction tree structures for mixed graphical models and hierarchical models.

Keywords

D-separation tree, decomposition, junction tree, separation tree, structural learning

2010 Mathematics Subject Classification

Primary 05C65. Secondary 62H05.

Full Text (PDF format)