Heuristics for the Extraction of Rules from Discrete-Time Recurrent Neural Networks

C. W. Omlin, C. L. Giles, C. B. Miller

Research output: Chapter in Book/Report/Conference proceedingConference contribution

19 Scopus citations

Abstract

Discrete recurrent neural networks can learn to correctly classify long strings of a regular language when trained on a small finite set of positive and negative example strings [Giles 90]. Rules defining the learned grammar can be extracted from networks by applying clustering heuristics in the output space of recurrent state neurons. We give empirical evidence that there exists a correlation between the generalization performance of recurrent neural networks for regular language recognition and the rules that can be extracted from a neural network. We present a heuristic that permits us to extract good rules from trained networks and test our method on networks which are trained to recognize a simple regular language.

Original languageEnglish (US)
Title of host publicationProceedings - 1992 International Joint Conference on Neural Networks, IJCNN 1992
PublisherInstitute of Electrical and Electronics Engineers Inc.
Pages33-38
Number of pages6
ISBN (Electronic)0780305590
DOIs
StatePublished - 1992
Event1992 International Joint Conference on Neural Networks, IJCNN 1992 - Baltimore, United States
Duration: Jun 7 1992Jun 11 1992

Publication series

NameProceedings of the International Joint Conference on Neural Networks
Volume1

Conference

Conference1992 International Joint Conference on Neural Networks, IJCNN 1992
Country/TerritoryUnited States
CityBaltimore
Period6/7/926/11/92

All Science Journal Classification (ASJC) codes

  • Software
  • Artificial Intelligence

Fingerprint

Dive into the research topics of 'Heuristics for the Extraction of Rules from Discrete-Time Recurrent Neural Networks'. Together they form a unique fingerprint.

Cite this