This matlab code implements Multi-Set Cannonical Correlation Analysis as explained in the first reference below. It does make a few additions to that simple approach described there. First, it adds regularization by reducing dimensions with PCA in each dataset. This is usefull if the data is noisy or ill-conditioned, i.e. you have lots of dimensions and not as many samples to estimate correlations reliably. Second, it computes the least-squares estimat of the inverse mapping from the CCA subspace back to the original data (called 'forward model' here). These additions are explained in the second reference.