一个基于张量火车分解的张量填充方法及在图像恢复中的应用

低秩张量填充在数据恢复中有广泛应用, 基于张量火车(TT) 分解的张量填充模型在彩色图像和视频以及互联网数据恢复中应用效果良好。本文提出一个基于三阶张量TT分解的填充模型。在模型中, 引入稀疏正则项与时空正则项, 分别刻画核张量的稀疏性和数据固有的块相似性。根据问题的结构特点, 引入辅助变量将原模型等价转化成可分离形式, 并采用临近交替极小化(PAM) 与交替方向乘子法(ADMM) 相结合的方法求解模型。数值实验表明, 两正则项的引入有利于提高数据恢复的稳定性和实际效果, 所提出方法优于其他方法。在采样率较低或图像出现结构性缺失时, 其方法效果较为显著。     

张量方法在信赖域算法中的应用

信赖域算法是求解无约束优化问题的一种有效的算法.对于该算法的子问题,本文将原来目标函数的二次模型扩展成四次张量模型,提出了一个带信赖域约束的四次张量模型优化问题的求解算法.该方法的最大特点是:不仅在张量模型的非稳定点可以得到下降方向及相应的迭代步长,而且在非局部极小值点的稳定点也可以得到下降方向及相应的迭代步长,从而在算法产生的迭代点列中存在一个子列收敛到信赖域子问题的局部极小值点.     

一种正则化非负张量分解算法及其新的有效加速策略

非负张量分解优化模型在高维图像处理与数据分析中占有重要地位.本文聚焦超光谱图像重构问题,提出一种正则化非负张量分解算法,然后给出三种新的有效加速策略,分别为分层降维循环迭代、误差校正以及“指数保号性”策略.利用所提出的这些加速策略对算法求解效率进行综合提升与改进.最后,通过数值测试来验证本文所提出的算法与加速策略的可行性与实用性.     

基于零模正则的神经网络剪枝方法

本文提出一种有效的神经网络剪枝方法。该方法对神经网络训练模型引入零模正则项来促使模型权重稀疏,并通过删减取值为零的权重来压缩模型。对所提出的零模正则神经网络训练模型,文中通过建立其等价MPEC形式的全局精确罚得到其等价的局部Lipschitz代理,然后通过用交替方向乘子法求解该Lipschitz代理模型对网络进行训练、剪枝。最后,对MLP和LeNet-5网络模型进行测试,分别在误差2.2%和1%下,取得97.43%和99.50%的稀疏度,达到很好的剪枝效果。     

低秩稀疏矩阵恢复的快速非单调交替极小化方法

针对低秩稀疏矩阵恢复问题的一个非凸优化模型,本文提出了一种快速非单调交替极小化方法.主要思想是对低秩矩阵部分采用交替极小化方法,对稀疏矩阵部分采用非单调线搜索技术来分别进行迭代更新.非单调线搜索技术是将单步下降放宽为多步下降,从而提高了计算效率.文中还给出了新算法的收敛性分析.最后,通过数值实验的比较表明,矩阵恢复的非单调交替极小化方法比原单调类方法更有效.     

图像恢复的一种快速迭代正则化方法

本文研究了将图像恢复问题转化为大型的线性不适定问题的求解.利用由Landweber迭代正则化方法改进所得到的快速收敛的迭代正则化方法,处理具有可分离点扩散函数的图像恢复问题.图像恢复实验表明该方法可大大提高收敛速度,且在计算中只需要较少的存储量.     

基于交替方向乘子法的大规模线性多商品流问题求解算法

企业的商品流通配送问题是典型的线性多商品流问题.由于经营规模的扩大和全球化运营模式的推行,企业所面临的问题规模正变得空前巨大,数据存储也越来越分散,传统方法已无法适应求解需求.本文基于交替方向乘子法(ADMM)的可分解性,提出一类随机ADMM算法,将大规模的问题分解成多个、规模比较小的问题,并采取随机顺序去求解这些小问题以及对偶问题,最终得到原问题的最优解.算法克服了ADMM的直接拓展求解多块问题时可能发散的缺点,并采用MnetGen生成器随机生成的多个规模不同的线性多商品流问题对算法进行了测试,验证了算法的有效性和高效的求解效率.     

融合欧拉弹性能及边缘检测项的图像分割模型

欧拉弹性能量作为正则项已经被成功应用于图像处理模型中.Egil Bae等人在L²欧拉弹性模型的基础上提出的ECV-L1模型具有偏向于分割凸轮廓的性质,但分割结果易受参数及弹性项的影响,使得分割结果不易紧贴目标物体边缘.在ECV-L1模型基础上引入边缘检测项提出新模型,并应用增广拉格朗日算法和交替方向乘子法对新模型进行数值求解.数值实验表明,新模型具有使分割结果保持为凸的性质,同时新模型的分割结果更易靠近目标物体边缘.     

随机Bregman ADMM及其在训练具有离散结构的支持向量机中的应用

针对具有多块可分结构的非凸优化问题提出了一类新的随机Bregman交替方向乘子法,在周期更新规则下, 证明了该算法的渐进收敛性; 在随机更新的规则下, 几乎确定的渐进收敛性得以证明。数值实验结果表明, 该算法可有效训练具有离散结构的支持向量机。     

交替方向法在保结构的有限元模型修正问题中的应用

本文应用交替方向法处理二次特征值反问题,并要求同时保持对称性,半正定性和稀疏性等结构性质.实验表明我们提出的方法是可行的.     

Robust tensor train component analysis

Robust Principal Component Analysis plays a key role in various fields such as image and video processing, data mining, and hyperspectral data analysis. In this paper, we study the problem of robust tensor train (TT) principal component analysis from partial observations, which aims to decompose a given tensor into the low TT rank and sparse components. The decomposition of the proposed model is used to find the hidden factors and help alleviate the curse of dimensionality via a set of connected low-rank tensors. A relaxation model is to minimize a weighted combination of the sum of nuclear norms of unfolding matrices of core tensors and the tensor ${?}_1$ norm. A proximal alternating direction method of multipliers is developed to solve the resulting model. Furthermore, we show that any cluster point of the convergent subsequence is a Karush-Kuhn-Tucker point of the proposed model under some conditions. Extensive numerical examples on both synthetic data and real-world datasets are presented to demonstrate the effectiveness of the proposed approach.     

Multi-view side information-incorporated tensor completion

Tensor completion originates in numerous applications where data utilized are of high dimensions and gathered from multiple sources or views. Existing methods merely incorporate the structure information, ignoring the fact that ubiquitous side information may be beneficial to estimate the missing entries from a partially observed tensor. Inspired by this, we formulate a sparse and low-rank tensor completion model named SLRMV. The ${\ell }_0$-norm instead of its relaxation is used in the objective function to constrain the sparseness of noise. The CP decomposition is used to decompose the high-quality tensor, based on which the combination of Schatten $p$-norm on each latent factor matrix is employed to characterize the low-rank tensor structure with high computation efficiency. Diverse similarity matrices for the same factor matrix are regarded as multi-view side information for guiding the tensor completion task. Although SLRMV is a nonconvex and discontinuous problem, the optimality analysis in terms of Karush-Kuhn-Tucker (KKT) conditions is accordingly proposed, based on which a hard-thresholding based alternating direction method of multipliers (HT-ADMM) is designed. Extensive experiments remarkably demonstrate the efficiency of SLRMV in tensor completion.     

A decomposition method for convex minimization problems and its application

1. IntroductionConsider the following special convex programming problem(P) adn{f(~) g(z); Ax = z},where f: Re - (--co, co] and g: Re - (--co, co] are closed proper convex functions andA is an m x n matrix. The Lagrangian for problem (P) is defined by L: Rad x Re x Re -- (~co, co] as follows:L(x, z, y) = f(x) g(z) (y, Ax ~ z), (1.1)where (., .) denotes the inner product in the general sense and 'y is the Lagrangian multiplierassociated with the constraint Ax = z. The augmented L…     

一类基于L0/1软间隔损失函数的低秩支持张量机

支持向量机作为基于向量空间的一种传统的机器学习方法,不能直接处理张量类型的数据,否则不仅破坏数据的空间结构,还会造成维度灾难及小样本问题。作为支持向量机的一种高阶推广,用于处理张量数据分类的支持张量机已经引起众多学者的关注,并应用于遥感成像、视频分析、金融、故障诊断等多个领域。与支持向量机类似,已有的支持张量机模型中采用的损失函数多为L_0/1函数的代理函数。将直接使用L_0/1这一本原函数作为损失函数,并利用张量数据的低秩性,建立针对二分类问题的低秩支持张量机模型。针对这一非凸非连续的张量优化问题,设计交替方向乘子法进行求解,并通过对模拟数据和真实数据进行数值实验,验证模型与算法的有效性。     

一种求解单调变分不等式的部分并行分裂LQP交替方向法

LQP交替方向法是求解可分离结构型单调变分不等式问题的一种非常有效的方法.它不仅可以充分地利用目标函数的可分结构,将原问题分解为多个更易求解的子问题,还更适合求解大规模问题.对于带有三个可分离算子的单调变分不等式问题,结合增广拉格朗日算法和LQP交替方向法提出了一种部分并行分裂LQP交替方向法,构造了新算法的两个下降方向,结合这两个下降方向得到了一个新的下降方向,沿着这个新的下降方向给出了最优步长.并在较弱的假设条件下,证明了新算法的全局收敛性.     

Digital watermarking scheme for colour remote sensing image based on quaternion wavelet transform and tensor decomposition

Digital watermarking is important for protecting the intellectual property of remote sensing images. Unlike watermarking in ordinary colour images, in colour remote sensing images, watermarking has an important requirement: robustness. In this paper, a robust nonblind watermarking scheme for colour remote sensing images, which considers both frequency and statistical pattern features, is constructed based on the quaternion wavelet transform (QWT) and tensor decomposition. Using the QWT, not only the abundant phase information can be used to preserve detailed host image features to improve the imperceptibility of the watermark, but also the frequency coefficients of the host image can provide a stable position to embed the watermark. To further strengthen the robustness, the global statistical feature structure acquired through the tensor Tucker decomposition is employed to distribute the watermark's energy among different colour bands. Because both the QWT frequency coefficients and the tensor decomposition global statistical feature structure are highly stable against external distortion, their integration yields the proposed scheme, which is robust to many image manipulations. A simulation experiment shows that our method can balance the trade‐off between imperceptibility and robustness and that it is more robust than the traditional QWT and discrete wavelet transform (DWT) methods under many different types of image manipulations.     

惩罚框架下求解广义Nash均衡问题的分解算法

广义Nash均衡问题(GNEP),是非合作博弈论中一类重要的问题,它在经济学、管理科学和交通规划等领域有着广泛的应用.本文主要提出一种新的惩罚算法来求解一般的广义Nash均衡问题,并根据罚函数的特殊结构,采用交替方向法求解子问题.在一定的条件下,本文证明新算法的全局收敛性.多个数值例子的试验结果表明算法是可行的,并且是有效的.     

A projection method and Kronecker product preconditioner for solving Sylvester tensor equations

The preconditioned iterative solvers for solving Sylvester tensor equations are considered in this paper.By fully exploiting the structure of the tensor equation,we propose a projection method based on the tensor format,which needs less flops and storage than the standard projection method.The structure of the coefficient matrices of the tensor equation is used to design the nearest Kronecker product(NKP) preconditioner,which is easy to construct and is able to accelerate the convergence of the iterative solver.Numerical experiments are presented to show good performance of the approaches.