A novel parallel clustering algorithm based on artificial immune network using nVidia CUDA framework

Abstract

In this paper, a novel parallel data clustering algorithm based on artificial immune network aiNet is proposed to improve its efficiency. In consideration of the restrictions of GPU, we carefully designed the data structure, algorithm flow and memory allocation strategy of the parallel algorithm and realized it using NVIDIA's CUDA framework. During the implementation, in order to fully explore its implicit parallelism, we allocated threads on GPU that represent the network cells of aiNet, and modified this algorithm to let those thread operations parallel during the clustering process. We calculated the affinity parallel, combined the random numbers with the local search algorithm to select the first n cell parallel, and did the network suppression parallel. Experimental results show that certain speedup can be obtained by using the proposed method. © 2011 Springer-Verlag.