1: \begin{abstract}

2: Canonical correlation analysis is a classical technique for exploring the relationship between two sets of variables. 

3: It has important applications in analyzing high dimensional datasets originated from genomics, imaging and other fields.

4: This paper considers

5: adaptive minimax and computationally tractable estimation of leading sparse canonical coefficient vectors in high dimensions. 

6: % Three intrinsically related problems are studied.

7:  % to fully address the topic. 

8: First, we establish separate

9: % establish the minimax rates of convergence under prediction loss. Separate 

10: minimax estimation rates for

11: canonical coefficient vectors of each set of random variables under 

12: no structural assumption on marginal covariance matrices.

13: Second, we propose a computationally feasible estimator to attain the optimal rates adaptively under an additional sample size condition.

14: Finally, we show that a sample size condition of this kind is needed for any randomized polynomial-time estimator to be consistent, assuming hardness of certain instances of the Planted Clique detection problem.  

15: The result is faithful to the Gaussian models used in the paper.

16:  % and is achieved by a novel reduction scheme.

17: As a byproduct, we obtain the first computational lower bounds for sparse PCA under the Gaussian single spiked covariance model.

18: % \nb{Referee 1 said it's too long}

19: \smallskip

20: 

21: \textbf{Keywords.} Convex programming, group-Lasso, Minimax rates, Computational complexity, Planted Clique,

22: Sparse CCA (SCCA), Sparse PCA (SPCA)

23: \end{abstract}

24: