多组变量典型相关分析的正交Carroll准则  

On orthogonal Carroll's criterion for multi-block data analysis

作　　者：林雪珍[1] 刘新国[1] 

机构地区：[1]中国海洋大学数学科学学院,山东青岛266100

出　　处：《应用数学与计算数学学报》2017年第2期176-190,共15页Communication on Applied Mathematics and Computation

基　　金：国家自然科学基金资助项目(11371333)

摘　　要：研究多组变量典型相关分析的正交Carroll准则.首先,把Carroll准则等价地转化为迹函数的极小值问题,由此导出最优性条件;然后,对Van de Velden给出的方法进行改进.为了提高算法的收敛速度和获得全局解的可能性,给出了初始点的选择策略;最后,分析了正交Carroll准则与其他一些问题之间的联系.We deal with the orthogonal Carroll＇s criterion for canonical correlation analysis of multi-block data. Firstly, we prove that this problem can be equivalently transformed into a minimization problem of trace function. Then, a modified version of the inexact alternating variable method is proposed. We suggest two starting point strategies to boost up the probability of finding a global solution. Numerical tests are presented to demonstrate the efficiency of the method proposed. Key words canonical correlation analysis of multiple sets of variables; Carroll＇s criterion; generalized asymmetric procrutes

关 键 词：多组变量典型相关分析 Carroll准则 广义非对称procrutes 

分 类 号：O212.4[理学—概率论与数理统计]