MOSAIC: A macro-connectionist expert systems generator

摘要

Artificial Intelligence (AI) and Artificial Neural Networks (ANNs) are two scientific disciplines which have concentrated tremendous efforts in the understanding and reproduction of human cognitive functions through simulation. Integration of concepts from both disciplines becomes increasingly necessary and natural. Expert systems involve several different aspects of cognition and, therefore, are interesting as a domain of integration. Concepts of each discipline must be selected in order to produce a synergism when they are merged together. At present, three approaches might be proposed: the interfacing approach, the extension approach, and the integration approach. In this article, we describe MOSAIC, the acronym for “Macro-connectionist Organization System for Artificial Intelligence Computation,” which corresponds to the integration approach and which presents a number of basic features: The numerical connectivity aspect, the autonomy of functional structured entities, and the recursive construction of assemblies of entities. These features allow MOSAIC to manage several inference strategies (forward chaining, backward chaining, and implicit deduction), to acquire knowledge explicitly or by a fast unsupervized learning from examples (Estimated Connection Weights Learning), to process uncertain and partial information, to integrate both declarative and procedural knowledge and to be an open system. Finally, medical expert system applications are described.