On "Mean-square reconstruction error"