A Framework for Combining Logical and Probabilistic Models

<p>Integrating the expressive power of first-order logic with the probabilistic reasoning power of Bayesian networks has attracted the interest of many researchers. We review many techniques for integration of first-order logic and Bayesian networks and propose a new framework that exploits the translation of logical knowledge into Bayesian networks. We present a new search algorithm, relation search, to carry out the translation. We prove that Bayesian networks (LBNs, Logical Bayesian Networks) constructed using relation search behave according to our logical intuition. We show by experiment that LBNs can achieve classification accuracy similar to FOIL (First-Order Inductive Learner) classifiers, thus suggesting that most of the probabilistic information contained in a dataset can be exploited by an LBN. As a byproduct of the experiment, a new technique is proposed for improving the accuracy of Bayesian network classifiers. In our framework, Bayesian network composition supported by an ontology can be used to combine LBNs with other Bayesian networks (PBNs, Probabilistic Bayesian Networks).</p>

Name thesis:degree_name

Ph.D.

Level thesis:degree_level

Campus Access Dissertation

Discipline thesis:degree_discipline

Computer Science and Engineering

Year

2011

Author dc:creator

• Wang, Jingsong

Contributors dc:contributor

• Marco Valtorta

• Computer Sciences
• Electrical and Computer Engineering
• Engineering
• Physical Sciences and Mathematics
• Bayesian network
• Classification
• First order logic
• Inductive logic programming

Statement dc:rights

• © 2011, Jingsong Wang

Repository record dc:identifier

https://scholarcommons.sc.edu/etd/804

OAI identifier oai:identifier

oai:scholarcommons.sc.edu:etd-1805