Layer Types: a Concept for Modeling Arrays of Communicating Objects
- Gerd Kock
- Cybernetics and Systems
The NeuroLution system consists of hardware and software components, which can be used as stand alone tools or in an integrated form. For many standard network types, simulators with comfortable graphical user interfaces are ooered. However, the system easily can be extended by user deened items. This exibility is achieved, because any simulator is characterized by two abstract deenitions: a network deenition written in the neural network description language CONNECT, and an interface deenition based on the GINA library of graphical user interface classes. Also, C++ classes generated from the CONNECT deenitions can be exported for the development of custom applications. The NeuroLution PCI board contains up to four SAND neurochips, and accelerates feedforward, RBF, and Kohonen networks. The peak performance is 800 MCPS. Using the driver's C interface, the board can be exploited as a stand alone tool. However, the integrated NeuroLu-tion system also ooers graphical simulators, as well as the underlying C++ classes, for exploiting the board. Beside presenting the NeuroLution system , the paper discusses requirements on neural network simulation tools and gives a short overview of the state of the art.