Integration of neural heuristics into knowledge-based inference

Abstract

Summary form only given, as follows. Analogies observed between relief networks and neural networks lead to the plausibility of introducing heuristics developed under the neural network approach into knowledge-based systems. An approach has been developed that maps a rule-based system into the neural architecture in both the structural and the behavioral aspects. Under this approach, the knowledge base and the inference engine are mapped into an entity called conceptualization, where a node represents a concept and a link represents a relation between two concepts. Inference in the conceptualization involves the propagation and combination of activations as well as maximizing information transmission through layers. Learning is based upon a mechanism called backpropagation, which allows proper modification of the connection strengths in order to be adapted to the environment. Finally, the validity of this approach has been demonstrated by experiments.<>