Exploring possible improvements on likelihood weighting by omitting superfluous information and incorporating backward sampling
Keywords
Loading...
Authors
Issue Date
2022-01-30
Language
en
Document type
Journal Title
Journal ISSN
Volume Title
Publisher
Title
ISSN
Volume
Issue
Startpage
Endpage
DOI
Abstract
Since exact inference in Bayesian networks is computationally intractable,
approximate inference is the more feasible approach to many
problems, in theory. This research focuses on proposed enhancements,
integrating backward simulation and exploiting the sufficient information
theorem, of stochastic simulation methods. These proposed improvements
are added to the likelihood weighting algorithm, and tested
on two large benchmark networks from the bnlearn repository. Evidence
is shown, that omitting superfluous information improves running
time, simply because less sampling has to be performed, regardless
of network structure, but does not have an effect on convergence.
Results on backward simulation however, are dubious.
Description
Citation
Supervisor
Faculty
Faculteit der Sociale Wetenschappen