## Using of GPUs for cluster analysis of large data by K-means method

Institute of mathematics and mathematical modeling NAS RK, Al-Farabi Kazakh National University, Almaty, Kazakhstan

arXiv:1402.3788 [cs.DC], (16 Feb 2014)

@article{2014arXiv1402.3788L,

author={Litvinenko}, N.},

title={"{Using of GPUs for cluster analysis of large data by K-means method}"},

journal={ArXiv e-prints},

archivePrefix={"arXiv"},

eprint={1402.3788},

primaryClass={"cs.DC"},

keywords={Computer Science – Distributed, Parallel, and Cluster Computing, 68W10, B.2.4, C.1.2, C.1.4},

year={2014},

month={feb},

adsurl={http://adsabs.harvard.edu/abs/2014arXiv1402.3788L},

adsnote={Provided by the SAO/NASA Astrophysics Data System}

}

This problem was solved within the framework of the grant project "Solving of problems of cluster analysis with application of parallel algorithms and cloud technologies" in the Institute of Mathematics and Mathematical Modelling in Almaty. The problem of cluster analysis for the large amount of data is very important in different areas of science – genetics, biology, sociology etc. At the same time, such statistical known packages as STATISTICA, STADIA, SYSTAT and others do not allow to solve large problems. The new algorithm that uses the high processing power of GPUs for solving clustering problems by the K-means method was developed. This algorithm is implemented as a C++ application in Microsoft Visual Studio 2010 with using the GPU Nvidia GeForce 660. The developed software package for solving clustering problems by the method of K – means with using GPUs allows us to handle up to 2 million records with number of features up to 25. The gain in the computing time is in factor 5. We plan to increase this factor up to 20-30 after improving the algorithms.

February 19, 2014 by hgpu