Article ID | Journal | Published Year | Pages | File Type |
---|---|---|---|---|
9653613 | Neurocomputing | 2005 | 21 Pages |
Abstract
In this paper, the REX method of fuzzy rule extraction from neural networks (NN) is presented. It is based on evolutionary algorithms. In the search process of the evolutionary algorithm, a set of rules describing the performance of the NN is found. An evolutionary algorithm is also responsible for obtaining proper fuzzy sets. Two approaches are compared, namely REX Pitt and REX Michigan. The main difference lies in the information contained in one chromosome. In REX Pitt, one individual represents a set of rules, while in REX Michigan it represents one rule. The obtained results are compared to other known methods. REX Pitt has very good efficiency, producing a small number of fuzzy rules, while REX Michigan creates more low quality rules.
Related Topics
Physical Sciences and Engineering
Computer Science
Artificial Intelligence
Authors
U. Markowska-Kaczmar, W. Trelak,