A layered information processing model for neural classification modules

We introduce the NEUROSIM development system in which a different approach to embed artificial neural network (ANN) components into technical systems is pursued. All system components are developed in an integrated environment and solve the global task using parallel processing techniques. The hierarchical system architecture with localized functionalities is based on the processing of extended patterns as the main information entities. We present an exemplary NEUROSIM implementation of the semantic layer object classifier, based on the principles of encapsulation and polymorphism. The described translation procedures allow the interpretation of output patterns in domain dependent terms. More sophisticated procedures can easily be incorporated.<
>