Some properties of LSQR for large sparse linear least squares problems

<正> It is well-known that many Krylov solvers for linear systems,eigenvalue problems,andsingular value decomposition problems have very simple and elegant formulas for residual norms.Theseformulas not only allow us to further understand the methods theoretically but also can be usedas cheap stopping criteria without forming approximate solutions and residuals at each step beforeconvergence takes place.LSQR for large sparse linear least squares problems is based on the Lanczosbidiagonalization process and is a Krylov solver.However,there has not yet been an analogouslyelegant formula for residual norms.This paper derives such kind of formula.In addition,the authorgets some other properties of LSQR and its mathematically equivalent CGLS.     

Partially function linear error-in-response models with validation data

This paper considers partial function linear models of the form Y =∫X(t)β(t)dt + g(T)with Y measured with error. The authors propose an estimation procedure when the basis functions are data driven, such as with functional principal components. Estimators of β(t) and g(t) with the primary data and validation data are presented and some asymptotic results are given. Finite sample properties are investigated through some simulation study and a real data application.     

最小二乘支持向量机的预优共轭梯度法

针对Suyken等人提出的最小二乘支持向量机的共轭梯度法在输入样本的个数较大时,需要求解高阶线性方程组这一缺陷,提出了一种新算法。该算法利用分块矩阵的思想将该高阶线性方程组系数矩阵降阶,为了提高收敛速度,克服数值的不稳定性,采用条件预优共轭梯度法求解低阶的线性方程组。通过仿真试验证明用本文方法训练最小二乘支持向量机比共轭梯度法的训练速度提高了将近一倍。     

Approximate conditional likelihood for generalized linear models with general missing data mechanism

The generalized linear model is an indispensable tool for analyzing non-Gaussian response data, with both canonical and non-canonical link functions comprehensively used. When missing values are present, many existing methods in the literature heavily depend on an unverifiable assumption of the missing data mechanism, and they fail when the assumption is violated. This paper proposes a missing data mechanism that is as generally applicable as possible, which includes both ignorable and nonignorable missing data cases, as well as both scenarios of missing values in response and covariate. Under this general missing data mechanism, the authors adopt an approximate conditional likelihood method to estimate unknown parameters. The authors rigorously establish the regularity conditions under which the unknown parameters are identifiable under the approximate conditional likelihood approach. For parameters that are identifiable, the authors prove the asymptotic normality of the estimators obtained by maximizing the approximate conditional likelihood. Some simulation studies are conducted to evaluate finite sample performance of the proposed estimators as well as estimators from some existing methods. Finally, the authors present a biomarker analysis in prostate cancer study to illustrate the proposed method.     

Empirical likelihood confidence region for parameters in linear errors-in-variables models with missing data

The multivariate linear errors-in-variables model when the regressors are missing at random in the sense of Rubin (1976) is considered in this paper. A constrained empirical likelihood confidence region for a parameter β ₀ in this model is proposed, which is constructed by combining the score function corresponding to the weighted squared orthogonal distance based on inverse probability with a constrained region of β ₀. It is shown that the empirical log-likelihood ratio at the true parameter converges to the standard chi-square distribution. Simulations show that the coverage rate of the proposed confidence region is closer to the nominal level and the length of confidence interval is narrower than those of the normal approximation of inverse probability weighted adjusted least square estimator in most cases. A real example is studied and the result supports the theory and simulation’s conclusion.     

New results about the relationship between optimally weighted least squares estimate and linear minimum variance estimate

The optimally weighted least squares estimate and the linear minimum variance estimate are two of the most popular estimation methods for a linear model. In this paper, the authors make a comprehensive discussion about the relationship between the two estimates. Firstly, the authors consider the classical linear model in which the coefficient matrix of the linear model is deterministic, and the necessary and sufficient condition for equivalence of the two estimates is derived. Moreover, under certain conditions on variance matrix invertibility, the two estimates can be identical provided that they use the same a priori information of the parameter being estimated. Secondly, the authors consider the linear model with random coefficient matrix which is called the extended linear model; under certain conditions on variance matrix invertibility, it is proved that the former outperforms the latter when using the same a priori information of the parameter. This research is supported in part by the National Natural Science Foundation of China under Grant Nos. 60232010, 60574032, and the Project 863 under Grant No. 2006AA12A104.     

Improved adaptive pruning algorithm for least squares support vector regression

As the solutions of the least squares support vector regression machine(LS-SVRM) are not sparse,it leads to slow prediction speed and limits its applications.The defects of the existing adaptive pruning algorithm for LS-SVRM are that the training speed is slow,and the generalization performance is not satisfactory,especially for large scale problems.Hence an improved algorithm is proposed.In order to accelerate the training speed,the pruned data point and fast leave-one-out error are employed to validate the temporary model obtained after decremental learning.The novel objective function in the termination condition which involves the whole constraints generated by all training data points and three pruning strategies are employed to improve the generalization performance.The effectiveness of the proposed algorithm is tested on six benchmark datasets.The sparse LS-SVRM model has a faster training speed and better generalization performance.     

Two-step estimators in partial linear models with missing response variables and error-prone covariates

A partial linear model with missing response variables and error-prone covariates is considered. The imputation approach is developed to estimate the regression coefficients and the nonparametric function. The proposed parametric estimators are shown to be asymptotically normal, and the estimators for the nonparametric part are proved to converge at an optimal rate. To construct confidence regions for the regression coefficients and the nonparametric function, respectively, the authors also propose the empirical-likelihood-based statistics and investigate the limit distributions of the empirical likelihood ratios. The simulation study is conducted to compare the finite sample behavior for the proposed estimators. An application to an AIDS dataset is illustrated.     

A two-stage estimation algorithm for a type of current status data

Success-failure life tests are widely used in reliability engineering research to evaluate the storage life of products,where the observed data are the current status data,usually summarized as the form of "binomial life data".For this type of data,this paper proposes a two-stage algorithm to estimate some commonly used lifetime distributions.This algorithm is automatic,intuitively appealing and simple to implement.Simulation studies show that compared with some existing methods,the proposed algorithm is more stable and efficient,especially in small sample situations,and it can also be extended to deal with some complicated lifetime distributions.     

最小二乘支持向量机的参数优选方法及应用

支持向量机是一种新型的学习方法,该方法以结构风险最小化原则取代传统机器学习中的经验风险最小化原则,在小样本的机器学习中显示出了优异的性能.传统的支持向量机是解凸二次规划问题,而最小二乘支持向量机是解等式线性方程,显得尤为方便,通过建立适当的性能指标,用遗传算法优化最小二乘支持向量机的有关参数,并在非线性经济系统中应用.用最小二乘支持向量机对非线性经济系统进行预测并与其它方法的预测结果比较,结果证明,该模型的预测精确度是令人满意的,文中提出的方法是可行的.     

基于最优LS-SVM的制导工具误差分离与折合

将最小二乘支持向量机方法应用于制导工具误差分离于折合。利用线性核函数获得了工具误差系数的估计,然后利用交叉验证技术推导了最小二乘支持向量机最优参数的选择准则。该准则的计算是基于模型求解的中间参数,所以并没有增加很多的计算量。最后根据六自由度弹道仿真软件进行了特殊弹道与全程弹道的仿真。仿真计算表明,与最小二乘和主成份方法相比,最优最小二乘支持向量机获得的误差系数估计与真值更加接近,折合得到的全程弹道遥外差更加准确。     

Smoothed estimator of quantile residual lifetime for right censored data

It is of great interest to estimate quantile residual lifetime in medical science and many other ?elds. In survival analysis, Kaplan-Meier(K-M) estimator has been widely used to estimate the survival distribution. However, it is well-known that the K-M estimator is not continuous, thus it can not always be used to calculate quantile residual lifetime. In this paper, the authors propose a kernel smoothing method to give an estimator of quantile residual lifetime. By using modern empirical process techniques, the consistency and the asymptotic normality of the proposed estimator are provided neatly.The authors also present the empirical small sample performances of the estimator. De?ciency is introduced to compare the performance of the proposed estimator with the naive unsmoothed estimator of the quantile residaul lifetime. Further simulation studies indicate that the proposed estimator performs very well.     

Model checking for general linear error-in-covariables model with validation data

In this paper, model checking problem is considered for general linear model when covariables are measured with error and an independent validation data set is available. Without assuming any error model structure between the true variable and the surrogate variable, the author first apply nonparametric method to model the relationship between the true variable and the surrogate variable with the help of the validation sample. Then the author construct a score-type test statistic through model adjustment. The large sample behaviors of the score-type test statistic are investigated. It is shown that the test is consistent and can detect the alternative hypothesis close to the null hypothesis at the rate n ^−r with 0 ≤ r ≤ 1/2. Simulation results indicate that the proposed method works well.     

基于PLS-结构方程的顾客满意度评价方法

传统结构方程模型虽然在分析中既能处理测量误差．又可描述潜变量之间的结构关系、但是，在求解结构方程模型的过程中严格依赖协方差矩阵，并且假设测量变量服从正态分布，由于顾客满意的频数分布常常为非正态分布，从而极大限制了该方法在顾客满意度研究方面的应用．本文研究基于PLS的结构方程模型，有效解决了在构造客户满意度模型中所遇到的变量非正态问题．并以肛结构方程原理为支撑，对商品房顾客满意度问题进行了应用研究．     

Existence of linear strategy equilibrium in insider trading with partial observations

This paper investigates a linear strategy equilibrium in insider trading in continuous time allowing market makers to know some information on the value of a stock. By the use of filtering theory, the authors prove that in a monopoly market, there exists a unique equilibrium of linear strategy of intensity in a closed form, such that the insider can make positive profits and at which, all of the private information on the value of the stock is released; and the more accurate the information on the value of the stock observed by the market makers, the less the positive profits are made by the insider, and even go to zero. However, there is no Nash equilibrium in a Cournot competition market between two insiders if they both adopt a linear strategy of intensity.     

条件线性高斯模型的Gauss Hermite filter-Kalman filter算法

针对条件线性高斯状态空间模型,提出了高斯厄密特滤波-卡尔曼滤波(Gauss Hermite filter-Kalmanfilter,GHF-KF)滤波算法。算法将模型中的条件线性状态方程代入观测方程,并融合线性状态的过程噪声和观测噪声,由GHF获得非线性状态的估计;再将非线性状态的估计均值代入线性状态方程与观测方程,由KF获得线性状态的估计;获得的非线性状态估计方差还用于修正由KF估计的线性状态,以提高精度。将GHF-KF算法应用于目标跟踪的仿真结果表明,与现有Rao-Blackwellized粒子滤波器RBPF相比,新方法在保证估计精度的同时,明显提高了实时性,计算时间仅约为RBPF的5%。     

Empirical likelihood for partially linear models under negatively associated errors

This paper proposes to use the blockwise empirical likelihood(EL) method to construct the confidence regions for the regression vector β in a partially linear model under negatively associated errors. It is shown that the blockwise EL ratio statistic for β is asymptotically χ~2 distributed. The result is used to obtain an EL-based confidence region for β. Results of a simulation study on the finite sample performance of the proposed confidence regions are reported.     

Further study strong consistency of M estimator in linear model for \tilde \rho-mixing random samples

The strong consistency of M estimators of the regression parameters in linear models for $\tilde \rho$ -mixing random errors under some mild conditions is established, which is an essential improvement over the relevant results in the literature on the moment conditions and mixing errors. Especially, Theorem of Wu (2005) is improved essentially on the moment conditions.     

Generalized F-test for high dimensional regression coefficients of partially linear models

This paper proposes a test procedure for testing the regression coefficients in high dimensional partially linear models based on the F-statistic. In the partially linear model, the authors first estimate the unknown nonlinear component by some nonparametric methods and then generalize the F-statistic to test the regression coefficients under some regular conditions. During this procedure, the estimation of the nonlinear component brings much challenge to explore the properties of generalized F-test. The authors obtain some asymptotic properties of the generalized F-test in more general cases,including the asymptotic normality and the power of this test with p/n ∈(0, 1) without normality assumption. The asymptotic result is general and by adding some constraint conditions we can obtain the similar conclusions in high dimensional linear models. Through simulation studies, the authors demonstrate good finite-sample performance of the proposed test in comparison with the theoretical results. The practical utility of our method is illustrated by a real data example.     

Semiparametric Bayesian inference for accelerated failure time models with errors-in-covariates and doubly censored data

This paper proposes a Bayesian semiparametric accelerated failure time model for doubly censored data with errors-in-covariates. The authors model the distributions of the unobserved covariates and the regression errors via the Dirichlet processes. Moreover, the authors extend the Bayesian Lasso approach to our semiparametric model for variable selection. The authors develop the Markov chain Monte Carlo strategies for posterior calculation. Simulation studies are conducted to show the performance of the proposed method. The authors also demonstrate the implementation of the method using analysis of PBC data and ACTG 175 data.