Semantic Networks and Neural Nets

摘要

Connected networks of nodes representing conceptual knowledge are widely employed in artificial intelligence and cognitive science. This report describes a direct way of realizing these semantic networks with neuron-like computing units. The proposed framework appears to offer several advantages over previous work. It obviates the need for a centralized knowledge base interpreter, thereby partially solving the problem of computational effectiveness and also embodies an evidential semantics for knowledge that provides a natural treatment of defaults, exceptions and inconsistent or conflicting information. The model employs a class of inference that may be characterized as working with a set of competing hypotheses, gathering evidence for each hypothesis and selecting the best among these. The resulting system has been simulated and is capable of supporting existing semantic network applications dealing with problems of recognition and recall in a uniform manner. (Author).