《双流形正则化的主成分分析算法.doc》由会员分享,可在线阅读,更多相关《双流形正则化的主成分分析算法.doc(2页珍藏版)》请在三一文库上搜索。
1、双流形正则化的主成分分析算法卢桂馥【期刊名称】小型微型计算机系统【年(卷),期】2016(037)012【摘要】 针对流形正则化的低秩矩阵分解算法 (Manifold Regularized Low- rank Matrix Factorization,MRLMF) 只考虑了样本间几何结构这一缺点 ,提出一 种双流形正则化的主成分分析算法 (Dual-manifold Regularized Principal Component Analysis,DMRPCA).DMRPCA 算法不仅利用样本间的局部几何 结构信息来构建 Laplacian 图,也利用特征间的局部几何结构来构建 Laplac
2、ian 图,并 将这两个 Laplacian 图作为 正则化项 引入到主成分 分析 (Principal Component Analysis,PCA) 算法目标函数中 .然后 ,设计了一种 DMRPCA 的求 解算法 .在实际数据库上的实验表明 ,DMRPCA 算法可以提高现有算法聚类的准 确率 ,从而验证了 DMRPCA 算法是可行的 .%To overcome the weakness of the manifold regularized low-rank matrix factorization (MRLMF) method,i.e.,it can only consider the
3、 structure of samples,we propose a novel algorithm called dual-manifold regularized principal component analysis (DMRPCA).DMRPCA not only uses the local geometrical information of samples to construct Laplacian graph,but uses the local geometrical information of features to construct Laplacian graph.Then the two Laplacian graph terms are integrated into the objective function of principal component analysis (PCA) as the regularization terms.A procedure for solving DMRPCA is also presented in the