1: \begin{abstract}

2: 	In this paper, we propose and study a distributed and secure algorithm for computing dominant 

3: 	(or truncated) singular value decompositions (SVD) of large and distributed data matrices. 

4: 	We consider the scenario where each node privately holds a subset of columns and only 

5: 	exchanges ``safe" information with other nodes in a collaborative effort to calculate a dominant 

6: 	SVD for the whole matrix.

7: 	In the framework of alternating direction methods of multipliers (ADMM), 

8: 	we propose a novel formulation for building consensus by equalizing 

9: 	subspaces spanned by splitting variables instead of equalizing themselves. 

10: 	This technique greatly relaxes feasibility restrictions and accelerates convergence significantly,

11: 	while at the same time yielding simple subproblems.

12: 	We design several algorithmic features, including a low-rank multiplier formula 

13: 	and mechanisms for controlling subproblem solution accuracies, 

14: 	to increase the algorithm's computational efficiency and reduce its communication overhead. 

15: 	More importantly, unlike many existing distributed or parallelized algorithms, 

16: 	our algorithm preserves the privacy of locally-held data; 

17: 	that is, none of the nodes can recover the data stored in another node through 

18: 	information exchanged during communications.

19: 	We present convergence analysis results, including a worst-case complexity 

20: 	estimate, and extensive experimental results indicating 

21: 	that the proposed algorithm, while safely guarding data privacy,

22: 	has a strong potential to deliver a cutting-edge performance, 

23: 	especially when communication costs are high.

24: \end{abstract}