The paper puts forward improved alternating least squares optimization for canonical correlation analysis by introducing a coupling between the subproblems that speeds up convergence, as well as an faster accelerated optimization. The reviewers found the contributions significant both in theory and practice and the paper sound and well-written.