A shift for accurate singular values¶in the differential qd algorithm

The differential qd algorithm with the incorporation of shifts computes the singular values of bidiagonal matrices preserving high accuracy [4]. In this paper we propose an algorithm for the shift which uses only the squares of the elements of the matrix, so that the complete algorithm is implemented in terms of squares until final blocks of sizes 1 × 1 or 2 × 2 are reached. This shift is always lower than the smallest singular value and yields quadratic convergence as in [4]. The algorithm which computes the shift uses some parameters also useful for convergence criteria. Moreover a slight modification in the single step is proposed which admits a larger domain of the matrices to which the differential qd algorithm can be applied, using IEEE arithmetic. Received: December 1998 / Accepted: January 2001     

Singular values and eigenvalues of non-hermitian block Toeplitz matrices

We study the asymptotic distribution of singular values and eigenvalues of non-Hermitian block Toeplitz matrices, generated by a matrix-valued periodic functionf, supposed to beL ². A distribution result concerning singular values, due to Avram, Parter and Tyrtyshnikov, is extended to the case wheref is matrix-valued, not necessarily square. Although Szegö's formula no longer holds iff is not Hermitian, we show that the same distribution formula still applies, virtually unchanged, provided test functions are supposed to be harmonic and not only continuous. Moreover, the class of harmonic test functions is proved to be optimal, as far as that formula is concerned.     

Computation of reduced-order models of multivariable systems by balanced truncation

Previous algorithms to obtain reduced-order models by balanced truncation in a single step either require a very specific way to solve a pair of Lyapunov equations or are suitable only for scalar or symmetric MIMO systems. In this paper, model reduction is revisited and an algorithm to obtain a reduced order model in one step only is proposed. As in the previous algorithms, the key point is to construct two rectangular matrices whose smaller dimensions are equal to the number of Hankel singular values to be kept in the lower model. Unlike the one-step algorithms available in the literature, the algorithm proposed here does not make any restriction to the way the Lyapunov equations necessary to obtain the controllability and observability gramians are solved. Furthermore, since the algorithm only relies on singular value decomposition, it is expected to be robust.     

System zeros analysis via the Moore-Penrose pseudoinverse and SVDof the first nonzero Markov parameter

A new characterization of system zeros of an arbitrary linear system described by a state-space model S(A, B, C, D) is presented. The transmission zeros are characterized as invariant zeros of an appropriate strictly proper system with a smaller number of inputs and outputs than the original system. The approach is based on singular value decomposition (SVD) of the first nonzero Markov parameter. This result together with characterization of invariant and decoupling zeros, based on the Moore-Penrose inverse of the first nonzero Markov parameter and the Kalman canonical decomposition theorem, provided in the first part of the paper yield a complete characterization of system zeros of an arbitrary multi-input/multi-output system     

广义分布参数系统的初值问题

本文定义了广义左逆、广义Ｆｏｕｒｉｅｒ变换，矩产方根等概念，论述了由带奇异系数矩耦合偏微分方程描述的广义分布参数系统，由广义Ｆｕｒｉｅｒ变换定理讨论了广义分布参数系统的初值问题，得到了该系统的解及其相容的初值条件。     

Generalized inverses and resolution in the solution of linear equations

A new generalized inverse of a singular and/or rectangular matrix is constructed by utilizing the Moore-Penrose generalized inverse. If this new inverse is used for solving a system of linear equations then it yields a solution vector with good resolution and small errors. Such solutions are particularly useful in remote sensing of atmospheric temperature profiles by satellites.     

CORDIC arithmetic for an SVD processor

Arithmetic issues in the calculation of the Singular Value Decomposition (SVD) are discussed. Traditional algorithms using hardware division and square root are replaced with the special-purpose CORDIC algorithms for computing vector rotations and inverse tangents. The CORDIC 2 × 2 SVD processor can be twice as fast as one assembled from traditional hardware units. A CORDIC SVD processor array is suitable for VLSI implementation and is important for use in real-time signal processing applications.     

Fast QR factorization of low-rank changes of Vandermonde-like matrices

This paper is concerned with the solution of linear systems with coefficient matrices which are Vandermonde-like matrices modified by adding low-rank corrections. Hereafter we refer to these matrices as modified Vandermonde-like matrices. The solution of modified Vandermonde-like linear systems arises in approximation theory both when we use Remez algorithms for finding minimax approximations and when we consider least squares problems with constraints. Our approach relies on the computation of an inverse QR factorization. More specifically, we show that some classical orthogonalization schemes for m×n, m≥n, Vandermonde-like matrices can be extended to compute efficiently an inverse QR factorization of modified Vandermonde-like matrices. The resulting algorithm has a cost of O(mn) arithmetical operations. Moreover it requires O(m) storage since the matrices Q and R are not stored. Received: January 1997 / Accepted: November 1997     

基于ELM-PNN算法的第24周太阳黑子预测预报

为了提高太阳黑子预测预报的精度,提出固定型极限学习过程神经网络(FELM-PNN)和增量型极限学习过程神经网络(IELM-PNN)两种学习算法.FELM-PNN的隐层节点数目固定,使用SVD求解隐层输出矩阵的Moore-Penrose广义逆,通过最小二乘法计算隐层输出权值;IELM-PNN逐次增加隐层节点,根据隐层输出矩阵和网络误差计算增加节点的输出权值.通过Henon时间序列预测验证了两种方法的有效性,并实际应用于第24周太阳黑子平滑月均值的中长期预测预报中.实验结果表明,两种方法的预测精度均有一定程度的提高,IELM-PNN的训练收敛性优于FELM-PNN.     

The representations and computations of generalized inverses $$A^{(1)}_{T,S}$$, $$A^{(1,2)}_{T,S}$$AT,S(1,2) and the group inverse

In this paper, we derive novel representations of generalized inverses \(A^{(1)}_{T,S}\) and \(A^{(1,2)}_{T,S}\), which are much simpler than those introduced in Ben-Israel and Greville (Generalized inverses: theory and applications. Springer, New York, 2003). When \(A^{(1,2)}_{T,S}\) is applied to matrices of index one, a simple representation for the group inverse \(A_{g}\) is derived. Based on these representations, we derive various algorithms for computing \(A^{(1)}_{T,S}\), \(A^{(1,2)}_{T,S}\) and \(A_{g}\), respectively. Moreover, our methods can be achieved through Gauss–Jordan elimination and complexity analysis indicates that our method for computing the group inverse \(A_{g}\) is more efficient than the other existing methods in the literature for a large class of problems in the computational complexity sense. Finally, numerical experiments show that our method for the group inverse \(A_{g}\) has highest accuracy among all the existing methods in the literature and also has the lowest cost of CPU time when applied to symmetric matrices or matrices with high rank or small size matrices with low rank in practice.     

埃尔米特自反矩阵的广义逆特征值问题与最佳逼近问题

在振动控制中,通常用矩阵的逼近问题来校正刚度矩阵和质量矩阵,使得它们具有给定的谱约束条件.本文基于埃尔米特自反矩阵的表示定理,利用矩阵的拉直和Kronecker积,得到了埃尔米特自反矩阵广义逆特征值问题解的一般表达式.进一步,对任意给定的n阶复矩阵对,利用Moor-Penrose广义逆和逼近理论,得到了其相关最佳逼近问题解的表达式.     

Methods of computing weighted pseudoinverse matrices with singular weights

Three iterative processes are constructed and investigated for computing weighted pseudoinverse matrices with singular weights and ML-weighted pseudoinverse matrices. Two of them are based on the decompositions of the weighted pseudoinverse matrix with singular weights into matrix power series, and the third is a generalization of the Schulz method to nonsingular square matrices. Translated from Kibernetika i Sistemnyi Analiz, No. 5, pp. 150–169, September–October, 1999.     

SPN分组密码中最优扩散层的构造与验证

SPN结构中的扩散层往往是矢量空间GF(2m)n上的线性变换,它与n阶矩阵在确定基 下一一对应。分支数B=n+1的扩散层是最优的,其充分必要条件是:对应矩阵的任意k阶子阵均 为非奇异。设计了构造最优SPN线性层的算法,并给出了线性变换最优扩散特性的验证算法。最 后,给出GF(28)8上一个最优线性变换及其验证结果的示例。     

On ordering fuzzy numbers

At the heart of many statistical processing algorithms lies the concept of ordering a set of crisp numbers, either according to their own values (“direct” sorting), or according to the values of a second set of numbers (“indirect” sorting). In this paper we show how the concept of direct and indirect sorting may be generalized to fuzzy numbers. We present two techniques for doing this: one is based on fuzzy permutation matrices and the other is based on Zadeh's extension principle. In the final section of the article we use the new direct and indirect fuzzy sorting techniques to define two new fuzzy number ordered weighted average (OWA) operators. © 2000 John Wiley & Sons, Inc.     

Gradient-type algorithms for partial singular value decomposition

It is often desirable to calculate only a few terms of the SVD expansion of a matrix, corresponding to the largest or smallest singular values. Two algorithms, based on gradient and conjugate gradient search, are proposed for this purpose. SVD is computed term by term in a decreasing or increasing order of singular values. The algorithms are simple to implement and are especially advantageous with large matrices.     

面向矩阵秩函数准确估计的自表示子空间聚类方法

传统子空间聚类方法通常使用矩阵核范数代替矩阵秩函数进行低秩矩阵恢复,然而在目标优化过程中主要关注低秩矩阵大奇异值的影响,容易导致矩阵秩估计不准确的问题。为此,在分析矩阵奇异值长尾分布特点基础上,提出使用基于截断Schatten-p范数的低秩子空间聚类模型。该模型充分考虑小奇异值对低秩矩阵恢复过程的贡献,利用小奇异值信息拟合矩阵奇异值的长尾分布,通过对矩阵秩函数进行准确估计以提升子空间聚类性能。实验结果表明,与现有加权核范数子空间聚类WNNM-LRR和近邻约束子空间聚类BDR算法相比,在Extended Yale B数据集上的聚类准确性分别提升了11%和8%,所提方法能够更好地拟合数据奇异值分布以及生成准确的相似度矩阵。     

Inverse dynamics of redundant manipulators using a minimum number of control forces

The number of the control actuators used by the inverse kinematics and dynamics algorithms that have been developed in the literature for generating redundant robot joint trajectories is equal to the number of the degrees of freedom of the manipulator. In this article, an inverse dynamics algorithm that performs the tasks using only a minimum number of actuators is proposed. The number of actuators is equal to the dimension of the task space, and the control forces are solved simultaneously with the corresponding system motion. It is shown that because all degrees of freedom are not actuated, the control forces may lose the ability to make an instantaneous effect on the end-effector acceleration at certain configurations, yielding the dynamical equation set of the system to be singular. The dynamical equations are modified in the neighborhood of the singular configurations by utilizing higher-order derivative information, so that the singularities in the numerical procedure are avoided. Asymptotically stable inverse dynamics closed-loop control in the presence of perturbations is also discussed. The algorithm is further generalized to closed chain manipulators. Three-link and two-link redundant planar manipulators are analyzed to illustrate the validity of the approach. © 2995 John Wiley & Sons, Inc.     

Generalized LQR control and Kalman filtering with relations to computations of inner-outer and spectral factorizations

We investigate the generalized linear quadratic regulator (LQR) control where the dimension of the control input is strictly greater than the dimension of the controlled output, and the weighting matrix on the control signal is singular. The dual problem is the generalized Kalman filtering where the dimension of the input noise process is strictly smaller than the dimension of the output measurement, and the covariance of the observation noise is singular. These two problems are intimately related to inner-outer factorizations for nonsquare stable transfer matrices with square inners of the smaller size. Such inner-outer factorizations are in turn related to spectral factorizations for power spectral density (PSD) matrices whose normal ranks are not full. We propose iterative algorithms and establish their convergence for inner-outer and spectral factorizations, which in turn solve the generalized LQR control and Kalman filtering.     

A new matrix test for randomness

Let ${{\mathcal S}}$ be one of the two multiplicative semigroups: M × M Boolean matrices, or the semigroup of M × M matrices over the field GF(2). Then for any matrix ${A\in {\mathcal S}}$ there exist two unique smallest numbers, namely the index and period k, d, such that A ^k  = A ^k+d . This fact allows us to form a new statistical test for randomness which we call the Semigroup Matrix Test. In this paper, we present details and results of our experiments for this test. We use Boolean matrices for M = 2, . . . , 5, and matrices over GF(2) of the size M = 2, . . . , 6. We also compare the results with the results obtained by the well-known Binary Matrix Rank Test.     

联合非下采样Contourlet变换与奇异值分解的多水印算法

为解决最近报道的Contourlet变换域基于奇异值分解的水印算法存在的高虚警率问题, 提出一种多水印算法。对Arnold置乱后的水印图像进行奇异值分解, 将其中一个正交矩阵嵌入原始图像非下采样Contourlet域的两个高频方向子带中, 并利用奇异值来调整原始图像非下采样Contourlet域剩余子带的系数矩阵, 通过逆变换得到含水印图像。抽取水印时首先计算从待检测图像抽出的正交矩阵和真实水印正交矩阵的相似度, 与阈值进行比较, 以决定抽取过程是否进行。由于非下采样Contourlet变换的高冗余性, 最终可抽取出多个水印图像。实验表明, 算法较好地克服了高虚警率问题。一系列的攻击实验证明算法具有较好的鲁棒性。