The principal component analysis can be carried out two ways. First the R-mode: R = K'K and the second is the Q-mode: Q = K K' where K is a data matrix centered by column or by row. The most commonly used method is the R-mode. It has been suggested that principal components computed from either the R-mode or the Q-mode may have the same interpretation. If this is true, then interpretation of the principal components could be put on a much more intuitive level in many applications. This will occur whenever one type of principal component is more intuitively related to the physical or natural system being studied than the other. The relationship between the principal components of the R-mode and the Q-mode have been investigated with the result that they show a perfect correlation between them. The conclusion that the principal components of the R-mode or the Q-mode have the same interpretation is established. An example is given to illustrate this work. The resulting interpretation is found to be the same as that obtained by Donald L. Phillips (1977) using different methods.

Linear factor models have attracted considerable interest over recent years especially in the econometrics literature. The intuitively appealing idea to explain a panel of economic variables by a few common factors is one of the reasons for their popularity. From a statistical viewpoint, the need to reduce the cross-section dimension to a much smaller factor space dimension is obvious considering the large data sets available in economics and finance. One of the characteristics of the traditional factor model is that the process is stationary in the time dimension. This appears restrictive, given the fact that over long time periods it is unlikely that e.g. factor loadings remain constant. For example, in the capital asset pricing model (CAPM) of Sharpe (1964) and Lintner (1965), typical empirical results show that factor loadings are time-varying, which in the CAPM is caused by time-varying second moments. In this thesis we generalize the tools of factor analysis for the study of stochastic processes whose behavior evolves over time. In particular, we introduce a new class of factor models with loadings that are allowed to be smooth functions of time. To estimate the resulting nonstationary factor model we generalize the properties of the principal components technique to the time-varying framework. We mainly consider separately two classes of Evolutionary Factor Models: Evolutionary Static Factor Models (Chapter 2) and Evolutionary Dynamic Factor Models (Chapter 3). In Chapter 2 we propose a new approximate factor model where the common components are static but nonstationary. The nonstationarity is introduced by the time-varying factor loadings, that are estimated by the eigenvectors of a nonparametrically estimated covariance matrix. Under simultaneous asymptotics (cross-section and time dimension go to infinity simultaneously), we give conditions for consistency of our estimators of the time varying covariance matrix, the loadings and the factors. This paper generalizes to the locally stationary case the results given by Bai (2003) in the stationary framework. A simulation study illustrates the performance of these estimators. The estimators proposed in Chapter 2 are based on a nonparametric estimator of the covariance matrix whose entries are computed with the same moothing parameter. This approach has the advantage of guaranteeing a positive definite estimator but it does not adapt to the different degree of smoothness of the different entries of the covariance matrix. In Chapter 5 we give an additional theoretical result which explains how to construct a positive definite estimate of the covariance matrix while while permitting different smoothing parameters. This estimator is based on the Cholesky decomposition of a pre-estimator of the covariance matrix. In Chapter 3 we introduce the dynamics in our modeling. This model generalizes the dynamic (but stationary) factor model of Forni et al. (2000), as well as the nonstationary (but static) factor model of Chapter 2. In the stationary (dynamic) case, Forni et al. (2000) show that the common components are estimated by the eigenvectors of a consistent estimator of the spectral density matrix, which is a matrix depending only on the frequency. In the evolutionary framework the dynamics of the model is explained by a time-varying spectral density matrix. This operator is a function of time as well as of the frequency. In this chapter we show that the common components of a locally stationary dynamic factor model can be estimated consistently by the eigenvectors of a consistent estimator of the time-varying spectral density matrix. In Chapter 4 we apply our theoretical results to real data and compare the performance of our approach with that based on standard techniques. Chapter 6 concludes and mention the main questions for future research.

Thesis (Ph.D)--Computing, Georgia Institute of Technology, 2009. / Committee Chair: Santosh Vempala; Committee Member: Adam Kalai; Committee Member: Haesun Park; Committee Member: Ravi Kannan; Committee Member: Vladimir Koltchinskii. Part of the SMARTech Electronic Thesis and Dissertation Collection.

In recent years, massive data sets are generated in many areas of science and business, and are gathered by using advanced data acquisition techniques. New approaches are therefore required to facilitate effective data management and data analysis in this big data era, especially to analyze multidimensional data for real-time applications. This thesis aims at developing generic and effective algorithms for compressing and recovering online multidimensional data, and applying such algorithms in image processing and other related areas. Since multidimensional data are usually represented by tensors, this research uses multilinear algebra as the mathematical foundation to facilitate development. After reviewing the techniques of singular value decomposition (SVD), principal component analysis (PCA) and tensor decomposition, this thesis deduces an effective multilinear principal component analysis (MPCA) method to process such data by seeking optimal orthogonal basis functions that map the original tensor space to a tensor subspace with minimal reconstruction error. Two real examples, 3D data compression for positron emission tomography (PET) and offline fabric defect detection, are used to illustrate the tensor decomposition method and the deduced MPCA method, respectively. Based on the deduced MPCA method, this research develops an incremental MPCA (IMPCA) algorithm which targets at compressing and recovering online tensor objects. To reduce computational complexity of the IMPCA algorithm, this research investigates the low-rank updates of singular values in the matrix and tensor domains, which leads to the development of a sequential low-rank update scheme similar to the sequential Karhunen-Loeve algorithm (SKL) for incremental matrix singular value decomposition, a sequential low-rank update scheme for incremental tensor decomposition, and a quick subspace tracking (QST) algorithm to further enhance the low-rank updates of singular values if the matrix is positive-symmetric definite. Although QST is slightly inferior to the SKL algorithm in terms of accuracy in estimating eigenvector and eigenvalue, the algorithm has lower computational complexity. Two fast incremental MPCA (IMPCA) algorithms are then developed by incorporating the SKL algorithm and the QST algorithm separately into the IMPCA algorithm. Results obtained from applying the developed IMPCA algorithms to detect anomalies from online multidimensional data in a number of numerical experiments, and to track and reconstruct the global surface temperature anomalies over the past several decades clearly confirm the excellent performance of the algorithms. This research also applies the developed IMPCA algorithms to solve an online fabric defect inspection problem. Unlike existing pixel-wise detection schemes, the developed algorithms employ a scanning window to extract tensor objects from fabric images, and to detect the occurrence of anomalies. The proposed method is unsupervised because no pre-training is needed. Two image processing techniques, selective local Gabor binary patterns (SLGBP) and multi-channel feature combination, are developed to accomplish the feature extraction of textile patterns and represent the features as tensor objects. Results of experiments conducted by using a real textile dataset confirm that the developed algorithms are comparable to existing supervised methods in terms of accuracy and computational complexity. A cost-effective parallel implementation scheme is developed to solve the problem in real-time. / published_or_final_version / Industrial and Manufacturing Systems Engineering / Doctoral / Doctor of Philosophy

published_or_final_version / Geography and Geology / Master / Master of Philosophy

Thesis (Ph.D.)--Baylor University, 2007. / Includes bibliographical references (p. 225-231).

Thesis (M.S.)--Ohio University, November, 2007. / Title from PDF t.p. Includes bibliographical references.

Diss. (sammanfattning) Uppsala : Uppsala universitet, 2005. / Härtill 6 uppsatser.

Thesis (Ph. D.)--University of Texas at Austin, 2008. / Vita. Includes bibliographical references.

Thesis (Ph. D.)--Virginia Polytechnic Institute and State University, 1992. / Vita. Abstract. Includes bibliographical references (leaves 105-107). Also available via the Internet.