high performance computing on graphics processing units: hgpu.org

The self organising map is a machine learning algorithm used to produce low dimensional representations of high dimensional data. While the process is becoming more and more useful with the rise of big data, it is hindered by the sheer amount of time the algorithm takes to run serially. This project produces a parallel version of the classical algorithm using OpenCL that is suited for use on parallel processor architectures. The use of the Manhattan distance metric is also explored as a replacement for the traditional Euclidean distance in an attempt to decrease run times. The output from the parallel program is compared to that of a widely used package, SOM PAK, to ensure validity. The parallel implementation is tested on a variety of architectures, including graphics hardware, to determine which parameters run times. A 10x speed up is demonstrated when compared with SOM PAK.