A new approach for computing canonical correlations and coordinates

Canonical correlation analysis (CCA) is an extremely useful technique in many applications that involve simultaneous analysis of a large number of variables of distinct types. In this paper, we present new methods of performing correlation analysis using gradient descent where canonical and variates and correlations are computed serially. The CCA is formulated as a solution of constrained and non-constrained optimization problems. Simulations are also provided to demonstrate the performance of the proposed techniques.