摘要
为恢复被混合噪声污染的低秩矩阵,提出了一种新的广义鲁棒主成分分析(GRPCA)算法。它通过最小化核范数、1范数和2,1范数的组合问题,从观测矩阵中分离出低秩部分和混合噪声部分,并用随机排序的交替方向乘子法求解。利用本文方法进行垃圾邮件分类的实验结果表明,与经典的主成分分析(PCA)和鲁棒主成分分析(RPCA)算法相比,本文方法可以有效提高垃圾邮件分类的精确度和稳定性。
A new generalized robust principal component analysis( GRPCA) algorithm is proposed in order to recover the low-rank matrix with mixed noise pollution. It separates the low-rank part and the mixed noise part from the observation matrix by minimizing the combination of the kernel norm,the 1 norm,and the 2,1 norm,and then solving by a randomly permuted alternating direction multiplier method. Using spam classification as an example and a comparison with the classic methods PCA and RPCA shows that this method can effectively improve the accuracy and robustness of spam classification.
作者
侯旭珂
杨宏伟
马方
赵丽娜
HOU XuKe;YANG HongWei;MA Fang;ZHAO LiNa(Faculty of Science;Center for Information Technology,Beijing University of Chemical Technology,Beijing 100029,China)
出处
《北京化工大学学报(自然科学版)》
CAS
CSCD
北大核心
2018年第4期82-85,共4页
Journal of Beijing University of Chemical Technology(Natural Science Edition)
基金
国家自然科学基金(11301021/11571031)
关键词
广义鲁棒主成分分析(GRPCA)
降维
k近邻(k
NN)
支持向量机(SVM)
generalized robust principal component analysis (GRPCA)
dimensionality reduction
k-nearest neigh-ber (kNN)
support vector machines (SVM)