Parallel Design And Implementation Of AP Clustering Algorithms Based On CUDA

The Affinity Propagation clustering aluorithm is a high-performance clustering algorithm published by Brendan J.Frey and Delbert Dueck in Science in 2007.It has been widely used in the fields of face image recognition,gene exon discovery.retrieval of the best air routes,handwritten zip codes,and facility location.However,as its application field needs to deal with the explosive growth of data volume,and its algorithm has higher time complexity,the time cost of AP clustering algorithm in dealing with large-scale different types of data sets is too large.Therefore,designing algorithms for rapid clustering of large-scale data has become one of the research hotspots.This paper achieves rapid clustering of large-scale data by improving optimization and parallel AP clustering algorithms.The principle analysis of AP clustering algorithm,the improvement of AP clustering algorithm and the parallel AP clustering algorithm based on CUDA platform are studied.This paper first analyzes the basic principles of AP clustering algorithm and the characteristics of AP clustering algorithm,finds its own limitations,and paves the way for the improvement and parallelism of AP clustering algorithm.Improving the optimized AP clustering algorithm is to increase the function of the AP clustering algorithm to better process the data set.Parallel improved AP clustering algorithm is the core content of this paper.Firstly,the AP clustering parallel algorithm based on CUDA platform is designed and found on the existing hardware devices.Then,the AP-CUDA clustering algorithm is analyzed by using the program analysis tool provided by CUDA.Find the optimal hardware settings to get the best performance and improve performance.Finally,serial AP clustering algorithm and AP-CUDA clustering algorithm are run on the hardware device to collect experimental data.The experimental data demonstrates that the improved optimized parallel AP-CUDA clustering algorithm has excellent performance when dealing with different types and larger data sets.