Parallel weighting K-means clustering algorithm based on graphics processing unit

In this paper, we consider the problem of clustering a large-scale data set. In particular, we present a Graphics Processing Unit (GPU) based Parallel Weighting k-means clustering algorithm (PW-kmeans) which enables us to utilize the parallel computing capability of GPUs to accelerate the running process of traditional weighting k-means algorithm. PW-kmeans works by transforming the operation of weighting k-means to the combination of multiplication, addition and element-wise operations among vectors or matrices. Since GPU has significant speed advantage to vector and matrix operations as opposed to CPU (Central Processing Unit), we develop parallel weighting k-means clustering algorithm with GPUs. Experimental results also demonstrate that the proposed technique outperforms current weighting kmeans algorithm with respect to the running speed.