Neural networks as virtual symbol systems

Symbolic knowledge based systems as well as neural networks offer several advantages but also suffer from several disadvantages if they are used as isolated systems. In consideration of this fact the main idea of this paper is to discuss several integration approaches and their opportunities of using the advantages of both paradigms in one system. The approach which is finally realized is called integration by cooperation and is presented in detail. Instead of realizing a hybrid system from scratch the suggested architecture integrates two competitive tools. The used expert system shell offers several symbolic knowledge representation schemes which are handled by different interpreters. Neural networks are treated in the suggested approach as an additional virtual symbolic knowledge representation and reasoning mechanism. From this point of view the described system offers a unified insight to hybrid problem solving. The architecture of the hybrid system as well as applications which are in progress are presented.