The effect of loop length and count on the approximation error and convergence speed of loopy belief propagation
Keywords
Loading...
Authors
Issue Date
2020-06-19
Language
en
Document type
Journal Title
Journal ISSN
Volume Title
Publisher
Title
ISSN
Volume
Issue
Startpage
Endpage
DOI
Abstract
Pearl’s classical belief propagation does not always converge or give correct
approximations of marginal probabilities when run on graphical models containing
loops. Earlier investigations have seen it working well on single-loop graphical
models as well as graphical models with multiple loops, but whether there is a direct
relation between the number and length of the loops in graphical model and
the accuracy and speed of convergence of LBP is not yet known. We found that
increasing the number of loops in the graphical model decreases the convergence
speed, while the accuracy of the marginal probabilities seems to decrease. Increasing
the loop length of a single loop also decreases the convergence speed, but the
accuracy of the marginal probabilities remains the same.
Keywords: loopy belief propagation; factor graph; inference; approximation;
loop length; loop count; cycle length; cycle count
Description
Citation
Supervisor
Faculty
Faculteit der Sociale Wetenschappen