Canonical Correlation Analysis:an Overview With Application To Learning Methods

We begin from the Canonical Correlation Analysis(CCA) that handles the samples with liner relationship, by leading into the kernel method, we generalize the tectonic processes of the Kernel Canonical Correlation Analysis(KCCA) that handles the samples with nonlinear relationship. In the process of the KCCA, we apply the regularization and incomplete Cholesky decomposition(ICD) in order to reduce the dimensionality of the kernel matrices and over the problem of over-fitting. Then, we introduce two kinds of kernel functions:Polynomial kernel and Gauss kernel. In the last part, we have the experiment of the correlation between two samples of show score based on the CCA and another experiment of handwritten numeral recognition based on the KCCA, verifying the handling effect of the CCA and KCCA within different samples and showing the effects of KCCA with different kernel functions.