A hardware supported system for Kohonen's Selforganizing Map

In this paper we present a system which enables easy and fast computation of Kohonen's selforganizing map (SOM). It is a hardware supported system consisting of different parts. A neural coprocessor (COKOS) is connected to a personal computer (PC) by a special, asynchronous interface. The neural coprocessor works vector-component-parallel and processing-element-serial and speeds up the performance of Kohonen's algorithm of the selforganizing map by magnitude. An userfriendly interface supports the preprocessing of the input data and the control and assessment of the learning process. 1 Introduction Large CPU-times for training large data sets to neural networks and the impossibility of real-time evaluation constitute a serious obstacle for practical applications of neural networks, especially software simulations on MIMD-computers, e.g. transputer networks [8], or on SIMD-computers (we are using MasPar with the language Parallaxis [1]), bring up the performance. For some general conce..