通过GPU加速数据挖掘的研究进展和实践  

Survey of research on GPU accelerated data mining and one case study

作　　者：戴春娥 陈维斌[1] 傅顺开[1] 李志强[1] 

机构地区：[1]华侨大学计算机科学与技术学院,福建厦门361021

出　　处：《计算机工程与应用》2015年第16期109-116,共8页Computer Engineering and Applications

基　　金：福建省绿色通信及其智能信息服务工程技术研究中心项目(No.2012H2002);厦门市科技计划基金(No.3505Z20133027);华侨大学人才引进科研基金(No.11Y0274);中央高校基本科研基金(No.11J0263)

摘　　要：将计算密度高的部分迁移到GPU上是加速经典数据挖掘算法的有效途径。首先介绍GPU特性和主要的GPU编程模型,随后针对数据挖掘主要任务类型分别介绍基于GPU加速的工作,包括分类、聚类、关联分析、时序分析和深度学习。最后分别基于CPU和GPU实现协同过滤推荐的两类经典算法,并基于经典的Movie Lens数据集的实验验证GPU对加速数据挖掘应用的显著效果,进一步了解GPU加速的工作原理和实际意义。Transferring the procedure involving dense computation to GPU is known as an effective way to accelerate the whole procedure of many classical data mining algorithms.In this paper,features of GPU as well as existing programming models of GPU are introduced firstly.The representative works of fundamental data mining tasks are covered respectively,including classification,clustering,association analysis,time series analysis and deep learning.Two classical algorithms of collaborative filtering are implemented on CPU and GPU,and experiments with Movie Lens data sets are conducted,which help to collect first-hand experience of applying GPU to accelerate the applications of data mining algorithms.

关 键 词：数据挖掘 GPU加速 并行计算 协同过滤 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]