A Perturbation Analysis of Low-Rank Matrix Recovery by Schatten p-Minimization

A number of previous papers have studied the problem of recovering low-rank matrices with noise, further combining the noisy and perturbed cases, we propose a nonconvex Schatten p-norm minimization method to deal with the recovery of fully perturbed low-rank matrices. By utilizing the p-null space property (p-NSP) and the p-restricted isometry property (p-RIP) of the matrix, sufficient conditions to ensure that the stable and accurate reconstruction for low-rank matrix in the case of full perturbation are derived, and two upper bound recovery error estimation ns are given. These estimations are characterized by two vital aspects, one involving the best r-approximation error and the other concerning the overall noise. Specifically, this paper obtains two new error upper bounds based on the fact that p-RIP and p-NSP are able to recover accurately and stably low-rank matrix, and to some extent improve the conditions corresponding to RIP.

Proximity point algorithm for low-rank matrix recovery from sparse noise corrupted data

The method of recovering a low-rank matrix with an unknown fraction whose entries are arbitrarily corrupted is known as the robust principal component analysis （RPCA）. This RPCA problem, under some conditions, can be exactly solved via convex optimization by minimizing a combination of the nuclear norm and the 11 norm. In this paper, an algorithm based on the Douglas-Rachford splitting method is proposed for solving the RPCA problem. First, the convex optimization problem is solved by canceling the constraint of the variables, and ~hen the proximity operators of the objective function are computed alternately. The new algorithm can exactly recover the low-rank and sparse components simultaneously, and it is proved to be convergent. Numerical simulations demonstrate the practical utility of the proposed algorithm.

Electrical Data Matrix Decomposition in Smart Grid

As the development of smart grid and energy internet, this leads to a significantincrease in the amount of data transmitted in real time. Due to the mismatch withcommunication networks that were not designed to carry high-speed and real time data,data losses and data quality degradation may happen constantly. For this problem,according to the strong spatial and temporal correlation of electricity data which isgenerated by human’s actions and feelings, we build a low-rank electricity data matrixwhere the row is time and the column is user. Inspired by matrix decomposition, we dividethe low-rank electricity data matrix into the multiply of two small matrices and use theknown data to approximate the low-rank electricity data matrix and recover the missedelectrical data. Based on the real electricity data, we analyze the low-rankness of theelectricity data matrix and perform the Matrix Decomposition-based method on the realdata. The experimental results verify the efficiency and efficiency of the proposed scheme.

Stable recovery of low-rank matrix via nonconvex Schatten p-minimization

In this paper, a sufficient condition is obtained to ensure the stable recovery(ε≠ 0) or exact recovery(ε = 0) of all r-rank matrices X ∈ Rm×nfrom b = A(X) + z via nonconvex Schatten p-minimization for anyδ4r∈ [3～(1/2))2, 1). Moreover, we determine the range of parameter p with any given δ4r∈ [(3～(1/2))/22, 1). In fact, for any given δ4r∈ [3～(1/2))2, 1), p ∈(0, 2(1- δ4r)] suffices for the stable recovery or exact recovery of all r-rank matrices.

Pairwise constraint propagation via low-rank matrix recovery

As a kind of weaker supervisory information, pairwise constraints can be exploited to guide the data analysis process, such as data clustering. This paper formulates pairwise constraint propagation, which aims to predict the large quantity of unknown constraints from scarce known constraints, as a low-rank matrix recovery(LMR) problem. Although recent advances in transductive learning based on matrix completion can be directly adopted to solve this problem, our work intends to develop a more general low-rank matrix recovery solution for pairwise constraint propagation, which not only completes the unknown entries in the constraint matrix but also removes the noise from the data matrix. The problem can be effectively solved using an augmented Lagrange multiplier method. Experimental results on constrained clustering tasks based on the propagated pairwise constraints have shown that our method can obtain more stable results than state-of-the-art algorithms,and outperform them.

基于判别低秩矩阵恢复和协同表示的遮挡人脸识别

针对训练样本和测试样本均受到严重的噪声污染的人脸识别问题,传统的子空间学习方法和经典的基于稀疏表示的分类(SRC)方法的识别性能都将急剧下降。另外,基于稀疏表示的方法也存在算法复杂度较高的问题。为了在一定程度上缓解上述问题,提出一种基于判别低秩矩阵恢复和协同表示的遮挡人脸识别方法。首先,低秩矩阵恢复可以有效地从被污损的训练样本中恢复出干净的、具备低秩结构的训练样本,而结构非相关性约束的引入可以有效提高恢复数据的鉴别能力。然后,通过学习原始污损数据与恢复出的低秩数据之间的低秩投影矩阵,将受污损的测试样本投影到相应的低维子空间,以修正污损测试样本。最后,利用协同表示的分类方法(CRC)对修正后的测试样本进行分类,获取最终的识别结果。在Extended Yale B和AR数据库上的实验结果表明,本文方法对遮挡人脸识别具有更好的识别性能。

Truncated sparse approximation property and truncated q-norm minimization

This paper considers approximately sparse signal and low-rank matrix’s recovery via truncated norm minimization minx∥xT∥q and minX∥XT∥Sq from noisy measurements.We first introduce truncated sparse approximation property,a more general robust null space property,and establish the stable recovery of signals and matrices under the truncated sparse approximation property.We also explore the relationship between the restricted isometry property and truncated sparse approximation property.And we also prove that if a measurement matrix A or linear map A satisfies truncated sparse approximation property of order k,then the first inequality in restricted isometry property of order k and of order 2k can hold for certain different constantsδk andδ2k,respectively.Last,we show that ifδs(k+|T^c|)<√(s-1)/s for some s≥4/3,then measurement matrix A and linear map A satisfy truncated sparse approximation property of order k.It should be pointed out that when Tc=Ф,our conclusion implies that sparse approximation property of order k is weaker than restricted isometry property of order sk.

The bounds of restricted isometry constants for low rank matrices recovery

This paper discusses conditions under which the solution of linear system with minimal Schatten-p norm, 0 〈 p ≤ 1, is also the lowest-rank solution of this linear system. To study this problem, an important tool is the restricted isometry constant （RIC）. Some papers provided the upper bounds of RIC to guarantee that the nuclear-norm minimization stably recovers a low-rank matrix. For example, Fazel improved the upper bounds to δ4Ar 〈 0.558 and δ3rA 〈 0.4721, respectively. Recently, the upper bounds of RIC can be improved to δ2rA 〈 0.307. In fact, by using some methods, the upper bounds of RIC can be improved to δ2tA 〈 0.4931 and δrA 〈 0.309. In this paper, we focus on the lower bounds of RIC, we show that there exists linear maps A with δ2rA 〉1√2 or δrA 〉 1/3 for which nuclear norm recovery fail on some matrix with rank at most r. These results indicate that there is only a little limited room for improving the upper bounds for δ2rA and δrA.Furthermore, we also discuss the upper bound of restricted isometry constant associated with linear maps A for Schatten p （0 〈 p 〈 1） quasi norm minimization problem.

SOLVING SYSTEMS OF QUADRATIC EQUATIONS VIA EXPONENTIAL-TYPE GRADIENT DESCENT ALGORITHM

We consider the rank minimization problem from quadratic measurements,i.e.,recovering a rank r matrix X 2 Rn×r from m scalar measurements yi=ai XX⊤ai,ai 2 Rn,i=1,...,m.Such problem arises in a variety of applications such as quadratic regression and quantum state tomography.We present a novel algorithm,which is termed exponential-type gradient descent algorithm,to minimize a non-convex objective function f(U)=14m Pm i=1(yi−a⊤i UU⊤ai)2.This algorithm starts with a careful initialization,and then refines this initial guess by iteratively applying exponential-type gradient descent.Particularly,we can obtain a good initial guess of X as long as the number of Gaussian random measurements is O(nr),and our iteration algorithm can converge linearly to the true X(up to an orthogonal matrix)with m=O(nr log(cr))Gaussian random measurements。