GDPC: A GPU-Accelerated Density Peaks Clustering Algorithm

Abstract

Density Peaks Clustering (DPC) is a recently proposed clustering algorithm that has distinctive advantages over existing clustering algorithms. However, DPC requires computing the distance between every pair of input points, therefore incurring quadratic computation overhead, which is prohibitive for large data sets. To address the efficiency problem of DPC, we propose to use GPU to accelerate DPC. We exploit a spatial index structure VP-Tree to help efficiently maintain the data points. We first propose a vectorized GPU-friendly VP-Tree structure, based on which we propose GDPC algorithm, where the density \(\rho \) and the dependent distance \(\delta \) can be efficiently computed by using GPU. Our results show that GDPC can achieve over 5.3–78.8\(\times \) acceleration compared to the state-of-the-art DPC implementations.