Randomized Generalized Singular Value Decomposition

The generalized singular value decomposition(GSVD)of two matrices with the same number of columns is a very useful tool in many practical applications.However,the GSVD may suffer from heavy computational time and memory requirement when the scale of the matrices is quite large.In this paper,we use random projections to capture the most of the action of the matrices and propose randomized algorithms for computing a low-rank approximation of the GSVD.Serval error bounds of the approximation are also presented for the proposed randomized algorithms.Finally,some experimental results show that the proposed randomized algorithms can achieve a good accuracy with less computational cost and storage requirement.

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.

Generalized Nonconvex Low-Rank Algorithm for Magnetic Resonance Imaging (MRI) Reconstruction

In recent years,utilizing the low-rank prior information to construct a signal from a small amount of measures has attracted much attention.In this paper,a generalized nonconvex low-rank(GNLR) algorithm for magnetic resonance imaging(MRI)reconstruction is proposed,which reconstructs the image from highly under-sampled k-space data.In the algorithm,the nonconvex surrogate function replacing the conventional nuclear norm is utilized to enhance the low-rank property inherent in the reconstructed image.An alternative direction multiplier method(ADMM) is applied to solving the resulting non-convex model.Extensive experimental results have demonstrated that the proposed method can consistently recover MRIs efficiently,and outperforms the current state-of-the-art approaches in terms of higher peak signal-to-noise ratio(PSNR) and lower high-frequency error norm(HFEN) values.

Parallel Active Subspace Decomposition for Tensor Robust Principal Component Analysis

Tensor robust principal component analysis has received a substantial amount of attention in various fields.Most existing methods,normally relying on tensor nuclear norm minimization,need to pay an expensive computational cost due to multiple singular value decompositions at each iteration.To overcome the drawback,we propose a scalable and efficient method,named parallel active subspace decomposition,which divides the unfolding along each mode of the tensor into a columnwise orthonormal matrix(active subspace)and another small-size matrix in parallel.Such a transformation leads to a nonconvex optimization problem in which the scale of nuclear norm minimization is generally much smaller than that in the original problem.We solve the optimization problem by an alternating direction method of multipliers and show that the iterates can be convergent within the given stopping criterion and the convergent solution is close to the global optimum solution within the prescribed bound.Experimental results are given to demonstrate that the performance of the proposed model is better than the state-of-the-art methods.

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.

Accurate simulations of pure-viscoacoustic wave propagation in tilted transversely isotropic media

Forward modeling of seismic wave propagation is crucial for the realization of reverse time migration(RTM) and full waveform inversion(FWI) in attenuating transversely isotropic media. To describe the attenuation and anisotropy properties of subsurface media, the pure-viscoacoustic anisotropic wave equations are established for wavefield simulations, because they can provide clear and stable wavefields. However, due to the use of several approximations in deriving the wave equation and the introduction of a fractional Laplacian approximation in solving the derived equation, the wavefields simulated by the previous pure-viscoacoustic tilted transversely isotropic(TTI) wave equations has low accuracy. To accurately simulate wavefields in media with velocity anisotropy and attenuation anisotropy, we first derive a new pure-viscoacoustic TTI wave equation from the exact complex-valued dispersion formula in viscoelastic vertical transversely isotropic(VTI) media. Then, we present the hybrid finite-difference and low-rank decomposition(HFDLRD) method to accurately solve our proposed pure-viscoacoustic TTI wave equation. Theoretical analysis and numerical examples suggest that our pure-viscoacoustic TTI wave equation has higher accuracy than previous pure-viscoacoustic TTI wave equations in describing q P-wave kinematic and attenuation characteristics. Additionally, the numerical experiment in a simple two-layer model shows that the HFDLRD technique outperforms the hybrid finite-difference and pseudo-spectral(HFDPS) method in terms of accuracy of wavefield modeling.

Fast nonnegative tensor ring decomposition based on the modulus method and low-rank approximation

Nonnegative tensor ring(NTR) decomposition is a powerful tool for capturing the significant features of tensor objects while preserving the multi-linear structure of tensor data. The existing algorithms rely on frequent reshaping and permutation operations in the optimization process and use a shrinking step size or projection techniques to ensure core tensor nonnegativity, which leads to a slow convergence rate, especially for large-scale problems. In this paper, we first propose an NTR algorithm based on the modulus method(NTR-MM), which constrains core tensor nonnegativity by modulus transformation. Second, a low-rank approximation(LRA) is introduced to NTR-MM(named LRA-NTR-MM), which not only reduces the computational complexity of NTR-MM significantly but also suppresses the noise. The simulation results demonstrate that the proposed LRA-NTR-MM algorithm achieves higher computational efficiency than the state-of-the-art algorithms while preserving the effectiveness of feature extraction.

求非凸二次规划全局最优解的分解线性化方法

对非凸二次规划(QP)问题提出新的确定性全局优化算法,该算法先对目标函数进行分解得到可分的等价问题,再根据相应函数的线性下估计建立原非凸二次规划的线性松弛规划,同时在分枝定界方法中使用区域删减准则来加速算法的收敛性.理论分析和数值计算表明提出的算法是收敛且有效的.

LMD与非凸罚最小化L_q正则子压缩传感的轴承振动信号重建

针对机械振动信号高速传输、大容量长期实时存储问题,提出一种局部均值分解(LMD)与非凸罚最小化Lq正则子压缩传感(CS)相结合的轴承故障振动信号重建方法。该方法利用振动系统信号采样、压缩合并进行的思想,首先通过LMD把振动信号分解为若干个不同频率分量的乘积函数平稳信号,对不同的频段分量寻求最佳的稀疏基,构建基于随机高斯矩阵的高度欠定方程;然后求解合适的压缩比,应用非凸罚最小化Lq正则子(q=0.5)算法重构,对所有重构信号组合得到原始振动信号。研究结果表明:LMD与非凸罚最小化Lq正则子压缩传感相结合的方法提高了轴承振动信号的重构精度,降低了重构计算复杂度,具有更高的处理速度和运行效率。

基于最优D.C.分解的单二次约束非凸二次规划精确算法

本文提出一种基于最优D.C.分解的单二次约束非凸二次规划精确算法.本文首先对非凸二次目标函数进行D.C.分解,然后对D.C.分解中凹的部分进行线性下逼近得到一个凸二次松弛问题.本文证明了最优D.C.分解可通过求解一个半定规划问题得到,而原问题的最优解可以通过计算最优凸二次松弛问题的满足某种互补条件的解得到.最后,本文报告了初步数值计算结果.

基于非凸矩阵分解的电网欺骗性数据注入攻击检测方法

电力系统状态估计中的量测数据容易受到欺骗性数据注入攻击的恶意篡改,使状态估计的稳定性受到影响。根据量测数据在连续时间段内的低维特性以及欺骗性数据攻击的稀疏特性,提出了一种基于非凸矩阵分解的电网欺骗性数据注入攻击检测方法。首先,将欺骗性数据注入攻击的检测问题视为稀疏低秩矩阵分解问题,并将分解问题转化为非凸优化问题,通过改进的交替方向乘子法求解此非凸问题,将受攻击的数据矩阵分解为正常量测矩阵和攻击矩阵;其次,利用分解出的攻击矩阵检测出欺骗性数据注入攻击的数值和位置,并以分解出的正常量测矩阵作为参考量测量,进行状态估计获得正确的状态变量;最后,通过IEEE-14节点系统分析了不同攻击幅值下的检测结果,验证了所提方法的准确性。

FAST ALGORITHMS FOR HIGHER-ORDER SINGULAR VALUE DECOMPOSITION FROM INCOMPLETE DATA

Higher-order singular value decomposition （HOSVD） is an efficient way for data reduction and also eliciting intrinsic structure of multi-dimensional array data. It has been used in many applications, and some of them involve incomplete data. To obtain HOSVD of the data with missing values, one can first impute the missing entries through a certain tensor completion method and then perform HOSVD to the reconstructed data. However, the two-step procedure can be inefficient and does not make reliable decomposition. In this paper, we formulate an incomplete HOSVD problem and combine the two steps into solving a single optimization problem, which simultaneously achieves imputation of missing values and also tensor decomposition. We also present one algorithm for solving the problem based on block coordinate update （BCU）. Global convergence of the algorithm is shown under mild assumptions and implies that of the popular higher-order orthogonality iteration （HOOI） method, and thus we, for the first time, give global convergence of HOOI. In addition, we compare the proposed method to state-of-the-art ones for solving incom- plete HOSVD and also low-rank tensor completion problems and demonstrate the superior performance of our method over other compared ones. Furthermore, we apply it to face recognition and MRI image reconstruction to show its practical performance.

基于DC分解的非凸二次规划SDP近似解

本文提出一类基于DC分解的非凸二次规划问题SDP松弛方法,并通过求解一个二阶锥问题得到原问题的近似最优解.我们首先对非凸二次目标函数进行DC分解,然后利用线性下逼近得到一个凸二次松弛问题,而最优的DC分解可通过求解一个SDP问题得到.数值试验表明,基于DC分解的SDP近似解平均优于经典SDP松弛和随机化方法产生的近似解。

Linear low-rank approximation and nonlinear dimensionality reduction

We present our recent work on both linear and nonlinear data reduction methods and algorithms: for the linear case we discuss results on structure analysis of SVD of columnpartitioned matrices and sparse low-rank approximation; for the nonlinear case we investigate methods for nonlinear dimensionality reduction and manifold learning. The problems we address have attracted great deal of interest in data mining and machine learning.

基于非凸低秩分解判别的叠加线性稀疏人脸识别

目的针对因采集的人脸图像样本受到污染而严重干扰人脸识别及训练样本较少(小样本)时会由于错误的稀疏系数导致性能急剧下降从而影响人脸识别的问题,提出了一种基于判别性非凸低秩矩阵分解的叠加线性稀疏表示算法。方法首先由γ范数取代传统核范数,克服了传统低秩矩阵分解方法求解核范数时因矩阵奇异值倍数缩放导致的识别误差问题;然后引入结构不相干判别项,以增加不同类低秩字典间的非相干性,达到抑制类内变化和去除类间相关性的目的;最后利用叠加线性稀疏表示方法完成分类。结果所提算法在AR人脸库中的识别率达到了98. 67±0. 57%,高于SRC(sparse representation-based classification)、ESRC(extended SRC)、RPCA(robust principal component analysis)+SRC、LRSI (low rank matrix decomposition with structural incoherence)、SLRC (superposed linear representation based classification)-l1等算法;同时,遮挡实验表明,算法对遮挡图像具有更好的鲁棒性,在不同遮挡比例下,相比其他算法均有更高的识别率。在CMU PIE人脸库中,对无遮挡图像添加0、10%、20%、30%、40%的椒盐噪声,算法识别率分别达到90. 1%、85. 5%、77. 8%、65. 3%和46. 1%,均高于其他算法。结论不同人脸库、不同比例遮挡和噪声的实验结果表明,所提算法针对人脸遮挡、表情和光照等噪声因素依然保持较高的识别率,鲁棒性更好。

非凸加权核范数及其在运动目标检测中的应用

目的近年来,低秩矩阵分解被越来越多的应用到运动目标检测中。但该类方法一般将矩阵秩函数松弛为矩阵核函数优化,导致背景恢复精度不高;并且没有考虑到前景目标的先验知识,即区域连续性。为此提出一种结合非凸加权核范数和前景目标区域连续性的目标检测算法。方法本文提出的运动目标检测模型以鲁棒主成分分析(RPCA)作为基础,在该基础上采用矩阵非凸核范数取代传统的核范数逼近矩阵低秩约束,并结合了前景目标区域连续性的先验知识。该方法恢复出的低秩矩阵即为背景图像矩阵,而稀疏大噪声矩阵则是前景目标位置矩阵。结果无论是在仿真数据集还是在真实数据集上,本文方法都能够取得比其他低秩类方法更好的效果。在不同数据集上,该方法相对于RPCA方法,前景目标检测性能提升25%左右,背景恢复误差降低0.5左右;而相对于DECOLOR方法,前景目标检测性能提升约2%左右,背景恢复误差降低0.2左右。结论矩阵秩函数的非凸松弛能够比凸松弛更准确的表征出低秩特征,从而在运动目标检测应用中更准确的恢复出背景。前景目标的区域连续性先验知识能够有效地过滤掉非目标大噪声产生的影响,使得较运动目标检测的精度得到大幅提高。因此,本文方法在动态纹理背景、光照渐变等较复杂场景中均能够较精确地检测出运动目标区域。但由于区域连续性的要求,本文方法对于小区域多目标的检测效果不甚理想。

Modeling the Correlations of Relations for Knowledge Graph Embedding

Knowledge graph embedding, which maps the entities and relations into low-dimensional vector spaces, has demonstrated its effectiveness in many tasks such as link prediction and relation extraction. Typical methods include TransE, TransH, and TransR. All these methods map different relations into the vector space separately and the intrinsic correlations of these relations are ignored. It is obvious that there exist some correlations among relations because different relations may connect to a common entity. For example, the triples （Steve Jobs, PlaceOfBrith, California） and （Apple Inc., Location, California） share the same entity California as their tail entity. We analyze the embedded relation matrices learned by TransE/TransH/TransR, and find that the correlations of relations do exist and they are showed as low-rank structure over the embedded relation matrix. It is natural to ask whether we can leverage these correlations to learn better embeddings for the entities and relations in a knowledge graph. In this paper, we propose to learn the embedded relation matrix by decomposing it as a product of two low-dimensional matrices, for characterizing the low-rank structure. The proposed method, called TransCoRe （Translation-Based Method via Modeling the Correlations of Relations）, learns the embeddings of entities and relations with translation-based framework. Experimental results based on the benchmark datasets of WordNet and Freebase demonstrate that our method outperforms the typical baselines on link prediction and triple classification tasks.

基于非凸加权核范数的目标视频帧选取

主要研究的是基于监控视频的显著前景目标帧选取问题．首先利用非凸加权核范数进行低秩背景估计．然后使用Markov随机场对稀疏前景目标位置矩阵进行估计，并提取出前景目标．最后定义显著前景目标区域，根据连通域面积大小，采用阈值法判断当前帧是否为显著目标帧，并采用F值来验证该模型的准确率．结果显示，模型在Campus、Curtain、Escalator、Fountain、Hall、Lobby和O伍ce这7个数据集上的F值均超过90％，说明该模型普遍具有较高的鲁棒性和准确性．

Mobile phone recognition method based on bilinear convolutional neural network

Model recognition of second-hand mobile phones has been considered as an essential process to improve the efficiency of phone recycling. However, due to the diversity of mobile phone appearances, it is difficult to realize accurate recognition. To solve this problem, a mobile phone recognition method based on bilinear-convolutional neural network(B-CNN) is proposed in this paper.First, a feature extraction model, based on B-CNN, is designed to adaptively extract local features from the images of secondhand mobile phones. Second, a joint loss function, constructed by center distance and softmax, is developed to reduce the interclass feature distance during the training process. Third, a parameter downscaling method, derived from the kernel discriminant analysis algorithm, is introduced to eliminate redundant features in B-CNN. Finally, the experimental results demonstrate that the B-CNN method can achieve higher accuracy than some existing methods.