A semantic network architecture for artificial intelligence processing

Abstract

A multiprocessor computer architecture for knowledge processing which is based on semantic network knowledge representation is described. The architecture has the capability of matching a semantic network query with the entire knowledge base in parallel; this in turn ensures that the best possible answer can be obtained. Applications in computer vision in general and in scene labeling in particular are described. It is shown that the introduction of weights to the links in the inference process gives some degree of fault tolerance and/or noise immunity. This architecture could have a significant impact on a wide range of artificial intelligence applications, such as knowledge bases, expert systems, high-level computer vision, and inference engines. The architecture might be implemented in wafer-scale integration technology. >