k-中心点聚合算法外文.docVIP

Efficient K-Means Clustering Using AcceleratedGraphics Processors S.A. Arul Shalom 1 , Manoranjan Dash 1 , and Minh Tue 2 1 School of Computer Engineering, Nanyang Technological University, 50 Nanyang Avenue, Singapore 2 NUS High School of Mathematics and Science, Clementi Avenue 3, Singapore {sall0001,asmdash}@.sg, h0630082@.sg Abstract. We exploit the parallel architecture of the Graphics Processing Unit(GPU) used in desktops to efficiently implement the traditional K-means algo-rithm. Our approach in clustering avoids the need for data and cluster informa-tion transfer between the GPU and CPU in between the iterations. In this paperwe present the novelties in our approach and techniques employed to representdata, compute distances, centroids and identify the cluster elements using the GPU. We measure performance using the metric: computational time per itera-tion. Our implementation of k-means clustering on an Nvidia 5900 graphics processor is 4 to 12 times faster than the CPU and 7 to 22 times faster on the Nvidia 8500 graphics processor for various data sizes. We also achieved 12 to 64 times speed gain on the 5900 and 20 to 140 times speed gains on the 8500 graphics processor in computational time per iteration for evaluations with various cluster sizes. Keywords: K-means clustering, GPGPU, Computational efficiency. 1 Introduction Commodity Graphics Processors in today’s PC world are highly parallel with extreme computational powers. Modern GPU is capable of processing tens of millions of ver- tices per second and rasterize hundreds of millions of fragments per second. The schematic in Figure 1 shows the vertex transformation and fragment texturing and coloring stages in a typical graphics pipeline. Due to the fact that these processors are economically affordable and programmable for implementing iterative algorithms there is high possibility that desktop computers with such GPUs will soon be capable of performing fast and efficient computing. 1.1 Graphics Pr