An "Artificial Expert"-Knowledge Acquisition via Neural Networks
Zhe , Ma.
- Publisher: Department of Automatic Control and Systems Engineering
arxiv: Computer Science::Neural and Evolutionary Computation
Artificial neural networks (ANN's) perform adaptive learning. This advantage can be used to solve knowledge acquisition bottle-neck in knowledge engineering by rule extraction from the ANN's. This paper proposes a rule extraction method combining both open-box (white-box) and black-box approaches to analyse a trained Multilayer Perceptron in order to extract general production rules accurately, abstractly and efficiently.