Han & Bonner map pairs of fMRI data from subs watching the same movie into orthogonal latent dims that maximize covariance. They find meaningful covariance up to 150 dims! Even dims of lower ranks show reliable individual differences correlated to behavior (movie recall).
