A comparison of feedback vertex set and loop cutset algorithms as preprocessing of conditioned polytree algorithm

Keywords

Loading...
Thumbnail Image

Issue Date

2022-01-29

Language

en

Document type

Journal Title

Journal ISSN

Volume Title

Publisher

Title

ISSN

Volume

Issue

Startpage

Endpage

DOI

Abstract

The poly-tree algorithm (Pearl’s algorithm) is an exact probabilistic inference method that updates messages from node to node until the marginal probability distribution of each node is consistent with the evidence [15]. The algorithm only works on singly connected networks. A possible way around this restriction is using conditioning algorithms to reduce the graph structure to obtain a singly connected network. To study the relative performance of exact versus heuristic approaches, I used one exact conditioning, loop cut-set, and one inexact conditioning method, feedback vertex set. Comparing the resulting performance measures and marginal probability distributions, I found that loop cut-set conditioning algorithm is preferred over the feedback vertex set conditioning in smaller networks and FVS had surprising accuracy in marginal dependencies despite a significant decrease in time required in larger networks. The varying number of loops merely affected the convergence rate performance of the loop cut-set in smaller networks. FVS heuristic conditioning recovered marginal probabilities better in larger networks compared to its performance on smaller networks. It performed faster and resulted in similar accuracy of the loop cut-set conditioning.

Description

Citation

Faculty

Faculteit der Sociale Wetenschappen