Automatic translation from an expert system to a neural network representation

The expert system is first translated into an intermediate, parallel, symbolic representation. That representation is then translated into a real-valued or binary neural network. These translation strategies were implemented in an automatic translation prototype, which was successfully applied to a number of expert system applications. A large majority of each expert system was translated; the untranslatable portions were nonetheless parallelized. The translator produced a highly parallel network, effectively examining all objects and most rules in parallel, and capable of running up to five orders of magnitude faster on existing neural hardware than on a 386 machine. The benefit of the translation process is that it most effectively combines the benefits of neural networks and expert systems, producing systems that are easy to design and understand because of their expert system origin and that are adaptable, fault tolerant, and parallel because of their neural network implementation