MAPPING OF SOM AND LVQ ALGORITHMS ON A TREE SHAPE PARALLEL COMPUTER-SYSTEM

Parallel mappings of Kohonen's self organizing map (SOM) and learning vector quantization (LVQ) algorithms are presented for a tree shape pa rallel computer system called TUTNC (Tampere University of Technology Neural Computer). The lattice of neurons in SOM is partitioned columnw ise to parallel processors in a neuron parallel manner. In addition, a n efficient method is presented for the neighborhood computation to ma ke the computation time independent of SOM size and processor count. T he tree shape architecture is shown to match well the requirements of mapped algorithms and their relations in such a prototype system TUTNC are studied, Performance has been measured for sample configurations and estimated for a larger system. Comparisons to other implementation s on Various platforms show, that good performance per processor has b een achieved.