LS_CCA: Least Squares formulation for Canonical Correlation Analysis

Liang Sun, Shuiwang Ji, and Jieping Ye

Introduction

LS_CCA is a Matlab implementation of the least squares formulation for Canonical Correlation Analysis (CCA). By constructing a specific target matrix from the label information, it is proved that CCA is equivalent to the LS_CCA formulation. Several extensions of LS_CCA based on regularization are also implemented, such as the sparse LS_CCA formulation using 1-norm regularization. This package also provides the implementations of CCA, kernel CCA as well as Orthonormalized Partial Least Squares (OPLS). Specifically, the implemented techniques include:

  1. CCA.
  2. Kernel CCA.
  3. Least squares.
  4. Kernel least squares.
  5. LS_CCA, the equivalent least squares formulation for CCA.
  6. KLS_CCA, the equivalent kernel least squares formulation for kernel CCA.
  7. Orthonormalized partial least squares (OPLS).

For all techniques, the regularization is supported. For least squares formulations, LS_CCA, and KLS_CCA, 1-norm and 2-norm regularization are supported.

Feedback

Please report any bug to Liang Sun (sun.liang@asu.edu). 

Download

LS_CCA is distributed for non-commercial use only.