A Comparison of Stochastic Processes and Artificial Neural Networks for Canonical Correlation Analysis

We have previously developed two artificial neural network methods [4], [3] of performing canonical correlation analysis (CCA). One of us [2] has recently developed a method of performing CCA using Gaussian processes; a second Bayesian method using latent variable models [1] has also recently been developed for CCA. No comparative results have been given for either of the Bayesian methods on real data sets. In this paper, we compare the accuracy of these four methods on a standard problem from [8].