Large scale K-means clustering using GPUs

Abstract The k -means algorithm is widely used for clustering, compressing, and summarizing vector data. We present a fast memory-efficient GPU-based exact -means, Asynchronous Selective Batched K (ASB -means). Unlike most algorithms that require loading the whole dataset onto GPU amount of memory required to run our can be chosen much smaller than size dataset. Thus, cluster datasets whose exceeds available memory. works in batched fashion applies triangle inequality each iteration omit data point if its membership assignment, i.e., it belongs to, remains unchanged, thus significantly reducing number points need transferred between CPU’s RAM GPU’s global enabling very efficiently process large datasets. Our substantially faster implementation standard even situations when application feasible because fits into Experiments show ASB up 15x times also outperforms NVIDIA’s open-source RAPIDS machine learning library on all experiments.