Using Iceberg Concept Lattices and Implications Rules to Extract Knowledge from Ann

Abstract

Nowadays, Artificial Neural Networks are being widely used in the representation of physical processes. Once trained, the networks are capable of solving unprecedented situations, keeping tolerable errors in their outputs. However, humans cannot assimilate the knowledge kept by these networks, since such knowledge is implicitly represented by their structure and connection weights. Recently, the FCANN method, based on Formal Concept Analysis, has been proposed as a new approach to extract, represent and understand the behavior of the processes based on rules. However, the extraction of those rules set is not an easy task. In this work, two main proposals to improve the FCANN method are presented and discussed: 1) the building of concept lattice using frequent item sets, which provides a threshold on the formal concepts number, and 2) the extraction of implications rules from the concept lattice, which provides clearer and more direct rules, thus facilitating the learning of the process by the user. As a c...