GPUMAFIA: Efficient Subspace Clustering with MAFIA on GPUs

摘要: Clustering, i.e., the identification of regions similar objects in a multi-dimensional data set, is standard method analytics with large variety applications. For high-dimensional data, subspace clustering can be used to find clusters among certain subset point dimensions and alleviate curse dimensionality. In this paper we focus on MAFIA algorithm using GPUs accelerate algorithm. We first present number algorithmic changes estimate their effect computational complexity These improve sequential version by 1---2 orders magnitude practical datasets while providing exactly same output. then GPU algorithm, which for typical provides further speedup over single CPU core or about an order multi-core CPU. believe that our faster implementation widens applicability clustering.