Parallelization of OBL based PSO K-means algorithm using OpenCL architecture
Qingyu Zhai, Dongfeng Yuan, Haixia Zhang, Kai Gao · 2014
To improve the searching performance to find better initial cluster centers and the calculating performance to process massive data in high dimensions, for PSO K-means, a brand new hybrid data clustering algorithm named Parallelization of OBL based PSO K-means Algorithm with the OpenCL Architecture (POPK) is introduced in this paper. In POPK, Opposition-based Learning (OBL) is applied to improve the global searching ability of Particle Swarm Optimization (PSO) in search of better initial centers of clusters for K-means while Open Computing Language (OpenCL) is introduced to parallelize the OBL-based PSO K-means and to enhance the calculating ability such that an obvious speed-up is obtained. Experimental results indicate that both effectiveness and efficiency of POPK is acceptably improved compared with standard PSO K-means.