To search, Click below search items.

 

All Published Papers Search Service

Title

Learning Model Transformation Rules from Examples: The GAILP System

Author

Hamdi A. Al-Jamimi and Moataz A. Ahmed

Citation

Vol. 19  No. 4  pp. 201-210

Abstract

Learning by examples refers to acquiring knowledge and experience to generalize theory from existing examples. Inductive logic programming (ILP) uses inductive inference to generate hypotheses from examples given with a background knowledge. ILP systems have been successfully applied in a number of real-world domains. Several ILP systems were introduced in the literature. Each system uses different search strategies and heuristics however, most systems employed a single predicate learning approach, which is not applicable in many learning problems. In this paper, we present GAILP, an ILP system that overcomes this limitation. GAILP employs genetic algorithms to discover various aspects of combinations to induce a set of hypotheses. It appraises such combinations in different ways to extract the most generic ones. The paper presents a thorough evaluation of the foundational aspects of the learning capability of GAILP. Two experiments were conducted to learn software model transformation rules. Experimental results reveal that GAILP is superior to a prominent ILP system, namely ALEPH, in different aspects and specifically in learning multi-predicates incrementally. We used a case study of tasks from the automated software engineering domain. The results obtained for the “class packaging” task showed that the accuracy of GAILP was 0.88 comparing with 0.83 achieved by ALEPH. Similarly, for “introducing Fa?ade interface” task, the accuracy obtained using GAILP and ALEPH were 0.90 and 0.66 respectively.

Keywords

Inductive logic programming inductive learning model transformation generalization.

URL

http://paper.ijcsns.org/07_book/201904/20190426.pdf