摘要
在分析传统集中式BP算法对海量高维数据分类不足的基础上,结合网格服务和粗糙集约简的思想,提出了基于网格服务的分布式BP分类算法(Distributed BP Classification algorithm based upon Grid Service,DBPC-GS)。仿真实验表明,粗糙集约简使得BP网络处理高维数据的复杂度大大降低;同时与传统集中式BP分类算法相比,DBPC-GS算法的平均耗时明显下降,CPU的负载也下降了约40%;且GBPC-GS算法仍然保持较高的分类精度。
This paper presents distributed BP classification algorithm based upon grid services (DBPC-GS), which combines grid services and reduction with rough set with distributed BP classification to solve problems of classifying mass and high-dimensional data by centralized BP algorithm. By simulation experiments, it' s showed that complexity which BP classify high-dimensional data is reduced greatly by reduction with rough set, meanwhile, the average consumptive time of DBPC-GS decreases apparently by contrast with concentrative algorithms, and CPU load is reduced about 40% with increments of grid codes. However, classification precision of DBPC-GS is on a level with concentrative BP classification algorithm.
出处
《南京邮电大学学报(自然科学版)》
EI
2008年第5期40-45,共6页
Journal of Nanjing University of Posts and Telecommunications:Natural Science Edition
基金
国家自然科学基金(60573141和60773041)
国家高技术研究发展计划(863计划)(2006AA01Z201、2006AA01Z439、2007AA01Z404、2007AA01Z478)
江苏省高技术研究计划(BG2006001)
南京市高科技项目(2007软资127)
现代通信国家重点实验室基金(9140C1105040805)
江苏省计算机信息处理技术重点实验室基金(kjs06006)资助项目
关键词
网格服务
BP神经网络
分类
粗糙集
约简
grid service
BP neural net
classification
rough set
reduction