Deep Tensor CCA for Multi-view Learning

We present Deep Tensor Canonical Correlation Analysis (DTCCA), a method to learn complex nonlinear transformations of multiple views (more than two) data such that the resulting representations are linearly correlated in high order. The high-order correlation given is modeled by covariance tensor, which different from most CCA formulations relying solely on pairwise correlations. Parameters each view jointly learned maximizing canonical correlation. To solve problem, we reformulate it as best sum rank-1 approximation, can be efficiently solved existing tensor decomposition method. DTCCA extension (TCCA) via deep networks. Comparing with kernel TCCA, not only deal arbitrary dimensions input data, but also does need maintain training for computing any point. Hence, unified model overcome scalable issue TCCA either high-dimensional multi-view or large amount views, and naturally extends learning representation. Extensive experiments four sets demonstrate effectiveness proposed