Inductive logic programming by instance patterns

• Published in: Lecture notes in computer science pp. 230 - 244
• Main Authors: Liu, Chongbing, Pontelli, Enrico
• Format: Conference Proceeding Book Chapter
• Language: English
• Published: Berlin, Heidelberg Springer-Verlag 2007; Springer Berlin / Heidelberg; Springer Berlin Heidelberg; Springer
• Series: ACM Other Conferences
• Subjects: Applied sciences; Computer science; control theory; systems; Computing methodologies; Computing methodologies > Artificial intelligence; Computing methodologies > Artificial intelligence > Knowledge representation and reasoning; Computing methodologies > Artificial intelligence > Knowledge representation and reasoning > Logic programming and answer set programming; Computing methodologies > Machine learning; Concept Instance; Exact sciences and technology; Inductive Logic Programming; Patterns; Programming languages; Software; Theory of computation; Theory of computation > Logic; Theory of computation > Logic > Constraint and logic programming; concept instance; patterns; inductive logic programming; Concept learning; Declarative language; Programming language; Theorem proving; Bottom up method; Top down method; Test coverage; Proof theory; Inductive logic programming
• ISBN: 3540696083; 9783540696087
• ISSN: 0302-9743; 1611-3349
• DOI: 10.1007/978-3-540-69611-7_15

Effectiveness and efficiency are two most important properties of ILP approaches. For both top-down and bottom-up search-based approaches, greater efficiency is usually gained at the expense of effectiveness. In this paper, we propose a bottom-up approach, called ILP by instance patterns, for the problem of concept learning in ILP. This approach is based on the observation that each example has its own pieces of description in the background knowledge, and the example together with these descriptions constitute a instance of the concept subject to learn. Our approach first captures the instance structures by patterns, then constructs the final theory purely from the patterns. On the effectiveness aspect, this approach does not assume determinacy of the learned concept. On the efficiency aspect, this approach is more efficient than existing ones due to its constructive nature, the fact that after the patterns are obtained, both the background and examples are not needed anymore, and the fact that it does not perform coverage test and needs no theorem prover.