Flexible Neuro-fuzzy Simulation Based on Abstract Model and Interface Definitions

This paper first presents an overview of neuro-fuzzy systems. Then a number of evaluation criteria for neuro-fuzzy simulation tools are established. An analysis of tools points out the difficulties in simultaneous fulfillment of this criteria, and leads to the development of design principles to overcome these difficulties. The basic idea is to implement simulators by two abstract descriptions a model and an interface definition. The MiND system, which has been built according to this idea, is then presented. On the one hand, MiND offers a comfortable graphical user interface, and on the other hand, it is flexible and extendible, and supports the development of custom applications. As a hardware component, the MiND system includes a Peripheral Compound Interconnect PCI neuroboard for the acceleration of neural networks. The paper closes with a discussion of the MiND system with respect to the evaluation criteria established at the beginning.