Neural Knowledge Processing in Expert Systems
Jiřı́ Šı́ma, Jiří Červenka · The MIT Press eBooks · 2000
In this chapter the knowledge-based neurocomputing will be applied to expert systems. Two main approaches to represent the knowledge base, namely the explicit and implicit representations will first be introduced and compared in rule-based and neural expert systems, respectively. Then, several possible integration strategies that make an effort to eliminate the drawbacks of both approaches in hybrid systems, will be surveyed. To illustrate the full power of knowledge-based neurocomputing, the main ideas of the prototypical, strictly neural expert system MACIE will be sketched. Here, a neural network is enriched by other functionalities to achieve all required features of expert systems. The neural knowledge processing will further be demonstrated on the system EXPSYS which exploits the powerful back-propagation learning to automatically create the knowledge base. In addition, EXPSYS introduces the interval neuron states to cope with incomplete information and it provides a simple expla...