| 基于MPI和OpenMP混合编程的非负矩阵分解并行算法 |
| |
| 引用本文： | 唐兵,Laurent BOBELIN,贺海武. 基于MPI和OpenMP混合编程的非负矩阵分解并行算法[J]. 计算机科学, 2017, 44(3): 51-54 |
| |
| 作者姓名： | 唐兵 Laurent BOBELIN 贺海武 |
| |
| 作者单位： | 湖南科技大学计算机科学与工程学院 湘潭411201,中国科学院计算机网络信息中心 北京100190,中国科学院计算机网络信息中心 北京100190 |
| |
| 基金项目： | 本文受中科院国际人才计划CAS PIFI(2016VTB028),湖南省自然科学基金(2015JJ3071)资助 |
| |
| 摘 要： | 非负矩阵分解(NMF)作为一种数据降维和特征提取的有效工具,已经在文本聚类、推荐系统等多个领域得到应用,但是其计算过程比较复杂。对此,提出一种基于MPI+OpenMP的混合层次化并行NMF方法,其充分利用基于MPI的消息传递模型和基于OpenMP的共享存储模型各自的优势,并基于多核节点集群进行测试。实验结果表明,所设计的并行NMF算法达到了较高的加速比,能有效处理高阶矩阵的非负分解,极大地提高了计算的效率。
|
| 关 键 词： | 非负矩阵分解 并行算法 MPI OpenMPI 可扩展 |
| 收稿时间： | 2015-12-28 |
| 修稿时间： | 2016-03-23 |
Parallel Algorithm of Nonnegative Matrix Factorization Based on Hybrid MPI and OpenMP Programming Model |
| |
| TANG Bing,Laurent BOBELIN and HE Hai-wu. Parallel Algorithm of Nonnegative Matrix Factorization Based on Hybrid MPI and OpenMP Programming Model[J]. Computer Science, 2017, 44(3): 51-54 |
| |
| Authors: | TANG Bing Laurent BOBELIN HE Hai-wu |
| |
| Affiliation: | School of Computer Science and Engineering,Hunan University of Science and Technology,Xiangtan 411201,China,Computer Network Information Center,Chinese Academy of Sciences,Beijing 100190,China and Computer Network Information Center,Chinese Academy of Sciences,Beijing 100190,China |
| |
| Abstract: | Nonnegative matrix factorization (NMF) has been introduced as an efficient way to reduce the complexity of data and extracting character,and it has also been applied to various fields,such as recommendations and text clustering.However,the computation process of NMF is quite complex.In order to solve this problem,a hybrid parallel hierar-chical NMF algorithm based on OpenMP and MPI was presented in this paper,which makes full use of the advantages of both MPI-based message passing model and OpenMP-based shared storage model.The new algorithm is evaluated in a multi-core cluster environment,and experimental results demonstrate that it can achieve a high speed-up,and can be used to deal with large-scale NMF with a high efficiency. |
| |
| Keywords: | Nonnegative matrix factorization Parallel algorithm MPI OpenMPI Scalability |
|
|
点击此处可从《计算机科学》下载全文 |
|
