1: \begin{abstract}

2: %

3: \large{

4: This paper considers the low-rank approximation problem with additional convex constraints, which is a non-convex problem.

5: %We formulate the generic non-convex problem and design a generic algorithm via the Alternating Direction Method of Multipliers (ADMM).

6: Two popular problems in current literature, Structured Low-Rank Approximation (\textbf{SLRA}) and Feasible Solution Recovery of SDP Relaxation (\textbf{FSR-SDPR}), can be viewed as two specific cases and many applications can be found. We show that the problem can be equivalently posed in an appropriate form for the Alternating Direction Method of Multipliers (ADMM) and a generic algorithm is derived based on that understanding. %To the best of our knowledge, this is the first attempt to use ADMM to solve this problem.

7: 

8: Different from the general non-convex problems, the subproblems in each step of ADMM can be solved efficiently in spite of their non-convexity. We verify its performance by extensive simulations based on synthetic and real-world data. Some preliminary theoretical results regarding to the convergence and feasibility are provided and more questions are left to be future work.

9: }

10: 

11: %We show that, for the non-convex problem, each sub-problem in ADMM can be solved efficiently and exactly. Structured Low-Rank Approximation (SLRA) and Feasible solutions Recovery of SDP Relaxation (FSR), which already attracted much attention in the literature, can be viewed as two specific cases and many applications can be found. %To the best of our knowledge, this is the first attempt to use ADMM to solve this non-convex problem. Some preliminary theoretical results are provided and more questions are left to be future work.}

12: \end{abstract}

13: