Minimax convergence rates for kernel CCA

Minimax convergence rates for kernel CCA
复制标题

DOI:
10.1016/j.jmva.2016.05.008
复制
发表时间:
2016-09
期刊:
J. Multivar. Anal.
影响因子:
--
通讯作者:
Zengyan Fan;H. Lian
Zengyan Fan;H. Lian
中科院分区:
其他
文献类型:
--
作者:
Zengyan Fan;H. Lian

文献摘要

被引文献

相似文献

核典型相关分析(CCA)的一致性已经建立,但其最佳收敛速度仍然未知。在本文中,我们推导出严格的上限和下限的核CCA的权函数的收敛速度。特别是最佳的收敛速度只取决于协方差算子的特征值的衰减率。
Consistency of kernel canonical correlation analysis (kernel CCA) has been established while its optimal convergence rate remains unknown. In this paper we derive rigorous upper and lower bounds for the convergence rate of the weight functions in kernel CCA. In particular the optimal convergence rate is shown to only depend on the rate of decay of the eigenvalues of the covariance operators.