Proximal iteratively reweighted algorithm for low-rank matrix recovery

This paper proposes a proximal iteratively reweighted algorithm to recover a low-rank matrix based on the weighted fixed point method. The weighted singular value thresholding problem gains a closed form solution because of the special properties of nonconvex surrogate functions. Besides, this study also has shown that the proximal iteratively reweighted algorithm lessens the objective function value monotonically, and any limit point is a stationary point theoretically.


Introduction
The low-rank matrix recovery problem has been a research hotpot recently [1,2], and it has a range of applications in many fields such as signal or image processing [3,4], subspace segmentation [5], collaborative filtering [6], and system identification [7]. Matrix rank minimization under affine equality constraints is generally formulated as follows: min X rank(X) s.t. A(X) = b, (1.1) where the linear map A : R m×n → R P and the vector b are known. Unfortunately, solving the above rank minimization problem (1.1) directly is an NPhard problem [8], thus this problem is computationally infeasible. Therefore, the convex relations of these methods have been proposed and studied in the literature. For example, Recht et al. [8] proposed a nuclear norm minimization method for the matrix reconstruction. The tightest convex relaxation of problem (1.1) is the following nuclear norm minimization problem: where X = r i=1 σ i (X) is the sum of all the singular values of X ∈ R m×n with rank(X) = r (without loss of generality, n ≤ m). It has been shown that problem (1.2) shares common solutions with problem (1.1) under some sufficient conditions (see, e.g., [8,9]).
However, the exact recovery of the low-rank matrix requires more measurements via nuclear norm minimization. Recently, some experimental observations and theoretical guarantees have shown the superiority of p quasi-norm minimization to 1 minimization in compressive sampling [10]. Therefore, the p quasi-norm minimization [11][12][13] was introduced instead of the nuclear norm minimization in order to give a better approximation to the original problem (1.1). Therefore, the p quasi-norm minimization can be formulated as where X p = ( r i=1 σ p i (X)) 1/p for some p ∈ (0, 1). However, in practice, the observed data in the low-rank matrix recovery problem may be contaminated with noise, namely b = AX + e, where e contains measurement errors dominated by certain normal distribution. In order to recover the low-rank matrix robustly, where · 2 is the 2 norm of vector and ε ≥ e 2 is some constant.
Under some conditions, problems (1.3) and (1.4) can be rewritten as the following unconstrained model: where τ > 0 is a given parameter. Since the above problem (1.5) is nonconvex and NPhard, thus the researchers throughout the world proposed and analyzed some iterative reweighted algorithms [13][14][15]. The key idea of the iterative reweighted technique is to solve a convex problem with a given weight at each iteration and update the weight at every turn.
Different from previous studies, based on the weighted fixed point method, this paper puts forward a proximal iteratively reweighted algorithm to recover a low-rank matrix.
Due to the special properties of nonconvex surrogate functions, the algorithm iteratively has a closed form solution to solve a weighted singular value thresholding problem. Also, in theory, this study has proved that the proximal iteratively reweighted algorithm decreases the objective function value monotonically, and any limit point is a stationary point.
The remainder of this paper is organized as follows. Section 2 introduces some notations and preliminary lemmas, and Section 3 describes the main results. The conclusion is followed in Section 4.

Preliminaries
Recently, Lai et al. [13] considered the following unconstrained problem: where I is the n × n identity matrix and ε > 0 is a smoothing parameter. By the definition in [13], we have is an absolutely symmetric function and p ∈ (0, 1), then ϕ = ψ • σ is subdifferentiable at matrix X ∈ R m×n and is a constant depending only on the value of σ i (X) for each i ∈ .
From Lemma 2.1, let m = n and the matrix Y be a semidefinite matrix, then Y = Y T and the subdifferentiable of the function Therefore, based on the definition of subgradient of the convex function, we have ). The inequality of (2.4) is equivalent to Then ϕ(Y k ) + G k , Y -Y k is used as a surrogate function of ϕ(Y ).

Main results
Let Y = X T X, then Y = V 2 V T can be obtained, where X = U V T with U ∈ R m×n , V ∈ R n×n , and = Diag{σ i (X)} ∈ R n×n , then σ i (Y ) = (σ i (X)) 2 . From (2.2), (2.3), and (2.5), In order to introduce the following lemma, the definitions of Lipschitz continuous of a function and the norm · F are given, namely a function is Lipschitz continuous with constant L if, for any x, y, |f (x)f (y)| ≤ L xy ; and the · F of a matrix X is defined as

thus the Lipschitz constant of the gradient ∇f (X) = A (A(X)b) is L(f ) = λ(A A), where λ(A A) is the maximum eigenvalue of A A.
By using (2.1), (2.3), (3.1), and (3.2), we update X k+1 by minimizing the sum of these two surrogate functions where ρ ≥ L(f ) 2 .

Lemma 3.2 If the function g(X)
= Q, X T X with X ∈ R m×n and Q ∈ R n×n , then the gradient of g(X) is ∇g(X) = 2XQ.
Proof Consider the auxiliary function θ : R → R, given by θ (t) = g(X + tY ), for any arbitrary matrix Y ∈ R m×n . From the basic calculus, it can be known that θ (0) = ∇g(X), Y . By the definition of the derivative of function, it follows that thus the gradient of g(X) is ∇g(X) = 2XQ.
Based on the above analysis, this paper proposes the following algorithm.

5:
Output low-rank matrix X k .
Proof Since X k+1 is the globally optimal solution of problem (3.3), and the zero matrix is contained in the subgradient with respect to X. That is, there exists a matrix X k+1 ∈ such that 2τ X k+1 W k + ∇f (X k ) + ρ(X k+1 -X k ) = 0. (3.5) By using the above equality of (3.4) and (3.5), we get Since the function W k , X T X is a convex function on X, thus and the above equality also can be rewritten as Then it follows from (3.6) and (3.7) that Let f (X) = 1 2 A(X)b 2 2 , and according to Lemma 3.1, can be obtained. Since the function tr((X T X + εI) p/2 ) is concave, and just like (3.1), then it can be obtained Now, combining (3.8), (3.9), and (3.10), we get Thus, F(X k ) is monotonically decreasing. Given the facts of all inequalities above for k ≥ 1, it can be obtained and from (3.11) it follows that Then, for k → ∞, (3.12) implies that lim k→∞ X k+1 -X k F = 0.
Since the objective function F(X) in problem (2.1) is nonnegative and satisfies then X k ∈ {X : 0 ≤ F(X) ≤ F(X 1 )} and the sequence {X k } is bounded. Therefore, the proof has been completed. .
Proof Since the sequence {X k } generated in Algorithm 1 is bounded, there exist an accumulation point X and a subsequence {X kj } such that lim j→∞ X kj = X . Assume that X kj is the solution of problem (3.3), it can be obtained 2τ X kj+1 W kj + ∇f (X kj ) + ρ(X kj+1 -X kj ) = 0.
Let j → ∞, according to Theorem 3.3, lim j→∞ X kj+1 -X kj F = 0 can be obtained. Hence, there exists the matrix By the above analysis, it can be known that τρX X T X + εI p/2-1 + ∇f (X ) = 0, then X is a stationary point of problem (2.1). Moreover, by using (3.11), for k = 1, 2, . . . , N , it can be obtained can be obtained, which completes the proof.

Conclusion
A proximal iteratively reweighted algorithm based on the weighted fixed point method for recovering a low-rank matrix problem has been presented in this paper. Due to the special properties of the nonconvex surrogate function, the algorithm in this study iteratively has a closed form solution to solving a weighted singular value thresholding problem. Finally, it has been proved that the algorithm can decrease the objective function value monotonically and any limit point is a stationary point.