Spectral factor model for time series learning

Today's computerized processes generatemassive amounts of streaming data.In many applications, data is collected for modeling the processes. The process model is hoped to drive objectives such as decision support, data visualization, business intelligence, automation and control, pattern recognition and classification, etc. However, we face significant challenges in data-driven modeling of processes. Apart from the errors, outliers and noise in the data measurements, the main challenge is due to a large dimensionality, which is the number of variables each data sample measures. The samples often form a long temporal sequence called a multivariate time series where any one sample is influenced by the others.We wish to build a model that will ensure robust generation, reviewing, and representation of new multivariate time series that are consistent with the underlying process.In this thesis, we adopt a modeling framework to extract characteristics from multivariate time series that correspond to dynamic variation-covariation common to the measured variables across all the samples. Those characteristics of a multivariate time series are named its 'commonalities' and a suitable measure for them is defined. What makes the multivariate time series model versatile is the assumption regarding the existence of a latent time series of known or presumed characteristics and much lower dimensionality than the measured time series; the result is the well-known 'dynamic factor model'.Original variants of existing methods for estimating the dynamic factor model are developed: The estimation is performed using the frequency-domain equivalent of the dynamic factor model named the 'spectral factor model'. To estimate the spectral factor model, ideas are sought from the asymptotic theory of spectral estimates. This theory is used to attain a probabilistic formulation, which provides maximum likelihood estimates for the spectral factor model parameters. Then, maximum likelihood parameters are developed with all the analysis entirely in the spectral-domain such that the dynamically transformed latent time series inherits the commonalities maximally.The main contribution of this thesis is a learning framework using the spectral factor model. We term learning as the ability of a computational model of a process to robustly characterize the data the process generates for purposes of pattern matching, classification and prediction. Hence, the spectral factor model could be claimed to have learned a multivariate time series if the latent time series when dynamically transformed extracts the commonalities reliably and maximally. The spectral factor model will be used for mainly two multivariate time series learning applications: First, real-world streaming datasets obtained from various processes are to be classified; in this exercise, human brain magnetoencephalography signals obtained during various cognitive and physical tasks are classified. Second, the commonalities are put to test by asking for reliable prediction of a multivariate time series given its past evolution; share prices in a portfolio are forecasted as part of this challenge.For both spectral factor modeling and learning, an analytical solution as well as an iterative solution are developed. While the analytical solution is based on low-rank approximation of the spectral density function, the iterative solution is based on the expectation-maximization algorithm. For the human brain signal classification exercise, a strategy for comparing similarities between the commonalities for various classes of multivariate time series processes is developed. For the share price prediction problem, a vector autoregressive model whose parameters are enriched with the maximum likelihood commonalities is designed. In both these learning problems, the spectral factor model gives commendable performance with respect to competing approaches.Les processus informatises actuels generent des quantites massives de flux de donnees. Dans nombre d'applications, ces flux de donnees sont collectees en vue de modeliser les processus. Les modeles de processus obtenus ont pour but la realisation d'objectifs tels que l'aide a la decision, la visualisation de donnees, l'informatique decisionnelle, l'automatisation et le controle, la reconnaissance de formes et la classification, etc. La modelisation de processus sur la base de donnees implique cependant de faire face a d’importants defis. Outre les erreurs, les donnees aberrantes et le bruit, le principal defi provient de la large dimensionnalite, i.e. du nombre de variables dans chaque echantillon de donnees mesurees. Les echantillons forment souvent une longue sequence temporelle appelee serie temporelle multivariee, ou chaque echantillon est influence par les autres. Notre objectif est de construire un modele robuste qui garantisse la generation, la revision et la representation de nouvelles series temporelles multivariees coherentes avec le processus sous-jacent.Dans cette these, nous adoptons un cadre de modelisation capable d’extraire, a partir de series temporelles multivariees, des caracteristiques correspondant a des variations - covariations dynamiques communes aux variables mesurees dans tous les echantillons. Ces caracteristiques sont appelees «points communs» et une mesure qui leur est appropriee est definie. Ce qui rend le modele de series temporelles multivariees polyvalent est l'hypothese relative a l'existence de series temporelles latentes de caracteristiques connues ou presumees et de dimensionnalite beaucoup plus faible que les series temporelles mesurees; le resultat est le bien connu «modele factoriel dynamique». Des variantes originales de methodes existantes pour estimer le modele factoriel dynamique sont developpees : l'estimation est realisee en utilisant l'equivalent du modele factoriel dynamique au niveau du domaine de frequence, designe comme le «modele factoriel spectral». Pour estimer le modele factoriel spectral, nous nous basons sur des idees relatives a la theorie des estimations spectrales. Cette theorie est utilisee pour aboutir a une formulation probabiliste, qui fournit des estimations de probabilite maximale pour les parametres du modele factoriel spectral. Des parametres de probabilite maximale sont alors developpes, en placant notre analyse entierement dans le domaine spectral, de facon a ce que les series temporelles latentes transformees dynamiquement heritent au maximum des points communs.La principale contribution de cette these consiste en un cadre d'apprentissage utilisant le modele factoriel spectral. Nous designons par apprentissage la capacite d'un modele de processus a caracteriser de facon robuste les donnees generees par le processus a des fins de filtrage par motif, classification et prediction. Dans ce contexte, le modele factoriel spectral est considere comme ayant appris une serie temporelle multivariee si la serie temporelle latente, une fois dynamiquement transformee, permet d'extraire les points communs de facon fiable et maximale. Le modele factoriel spectral sera utilise principalement pour deux applications d'apprentissage de series multivariees : en premier lieu, des ensembles de donnees sous forme de flux venant de differents processus du monde reel doivent etre classifies; lors de cet exercice, la classification porte sur des signaux magnetoencephalographiques obtenus chez l'homme au cours de differentes tâches physiques et cognitives; en second lieu, les points communs obtenus sont testes en demandant une prediction fiable d'une serie temporelle multivariee etant donnee l'evolution passee; les prix d'un portefeuille d'actions sont predits dans le cadre de ce defi.A la fois pour la modelisation et pour l'apprentissage factoriel spectral, une solution analytique aussi bien qu'une solution iterative sont developpees. Tandis que la solution analytique est basee sur une approximation de rang inferieur de la fonction de densite spectrale, la solution iterative est basee, quant a elle, sur l'algorithme de maximisation des attentes. Pour l'exercice de classification des signaux magnetoencephalographiques humains, une strategie de comparaison des similitudes entre les points communs des differentes classes de processus de series temporelles multivariees est developpee. Pour le probleme de prediction des prix des actions, un modele vectoriel autoregressif dont les parametres sont enrichis avec les points communs de probabilite maximale est concu. Dans ces deux problemes d’apprentissage, le modele factoriel spectral atteint des performances louables en regard d’approches concurrentes.

[1]  Wm. R. Wright General Intelligence, Objectively Determined and Measured. , 1905 .

[2]  R. F.,et al.  Mathematical Statistics , 1944, Nature.

[3]  N. L. Johnson,et al.  Multivariate Analysis , 1958, Nature.

[4]  Iver A. Lund,et al.  SOME OBJECTIVE SIX-HOUR PREDICTIONS PREPARED BY STATISTICAL METHODS , 1959 .

[5]  A. C. Young,et al.  Factor Analysis of the Electrocardiogram: Test of Electrocardiographic Theory Normal Hearts , 1960, Circulation research.

[6]  Richard Bellman,et al.  Adaptive Control Processes: A Guided Tour , 1961, The Mathematical Gazette.

[7]  N. R. Goodman Statistical analysis based on a certain multivariate complex Gaussian distribution , 1963 .

[8]  M. Nerlove SPECTRAL ANALYSIS OF SEASONAL ADJUSTMENT PROCEDURES , 1964 .

[9]  B. King Market and Industry Factors in Stock Price Behavior , 1966 .

[10]  David R. Cox,et al.  The Theory of Stochastic Processes , 1967, The Mathematical Gazette.

[11]  Thomas M. Cover,et al.  Estimation by the nearest neighbor rule , 1968, IEEE Trans. Inf. Theory.

[12]  E. B. Andersen,et al.  Modern factor analysis , 1961 .

[13]  E. Hannan,et al.  Spectral inference over narrow bands , 1971, Journal of Applied Probability.

[14]  G. Shilov,et al.  Linear Algebra , 1971 .

[15]  H. Tong,et al.  Applications of principal component analysis and factor analysis in the identification of multivariable systems , 1974 .

[16]  P. Lafrance,et al.  Digital filters , 1974, Proceedings of the IEEE.

[17]  Chris Chatfield,et al.  Introduction to Statistical Time Series. , 1976 .

[18]  Thomas J. Sargent,et al.  Business cycle modeling without pretending to have too much a priori economic theory , 1976 .

[19]  D. Rubin,et al.  Maximum likelihood from incomplete data via the EM - algorithm plus discussions on the paper , 1977 .

[20]  H. Theil Introduction to econometrics , 1978 .

[21]  F. Ahrabi Maximum likelihood estimation of the autoregressive coefficients and moving average covariances of vector autoregressive moving average models , 1979 .

[22]  Bruno O. Shubert,et al.  Random variables and stochastic processes , 1979 .

[23]  John Geweke,et al.  Maximum Likelihood "Confirmatory" Factor Analysis of Economic Time Series , 1981 .

[24]  G. Grimmett,et al.  Probability and random processes , 2002 .

[25]  Isaac Elishakoff,et al.  Probabilistic Theory of Structures , 1983 .

[26]  Peter C. M. Molenaar,et al.  A dynamic factor model for the analysis of multivariate time series , 1985 .

[27]  John G. Proakis,et al.  Probability, random variables and stochastic processes , 1985, IEEE Trans. Acoust. Speech Signal Process..

[28]  Jonathan D. Cryer,et al.  Time Series Analysis , 1986 .

[29]  Robert M. Gray,et al.  Probability, Random Processes, And Ergodic Properties , 1987 .

[30]  William L. Briggs,et al.  The DFT : An Owner's Manual for the Discrete Fourier Transform , 1987 .

[31]  Roman Kuc,et al.  Introduction to Digital Signal Processing , 2021, Digital Signal Processing.

[32]  John A. Pearce,et al.  A new window and comparison to standard windows , 1989, IEEE Trans. Acoust. Speech Signal Process..

[33]  John H. Karl,et al.  An introduction to digital signal processing , 1989 .

[34]  G. C. Wei,et al.  A Monte Carlo Implementation of the EM Algorithm and the Poor Man's Data Augmentation Algorithms , 1990 .

[35]  Muni S. Srivastava,et al.  Regression Analysis: Theory, Methods, and Applications , 1991 .

[36]  Richard A. Davis,et al.  Time Series: Theory and Methods , 2013 .

[37]  Adi Ben-Israel,et al.  On principal angles between subspaces in Rn , 1992 .

[38]  Roch Roy,et al.  ON THE INVERTIBILITY OF MULTIVARIATE LINEAR PROCESSES , 1993 .

[39]  Pierre Comon,et al.  Independent component analysis, A new concept? , 1994, Signal Process..

[40]  C C Wood,et al.  Mapping function in the human brain with magnetoencephalography, anatomical magnetic resonance imaging, and functional magnetic resonance imaging. , 1995, Journal of clinical neurophysiology : official publication of the American Electroencephalographic Society.

[41]  G. Reinsel Elements of Multivariate Time Series Analysis , 1995 .

[42]  J. A. Stewart,et al.  Nonlinear Time Series Analysis , 2015 .

[43]  F. Y. Edgeworth,et al.  The theory of statistics , 1996 .

[44]  Richard A. Davis,et al.  Introduction to time series and forecasting , 1998 .

[45]  P Bühlmann,et al.  What is a linear process? , 1996, Proceedings of the National Academy of Sciences of the United States of America.

[46]  James Demmel,et al.  Applied Numerical Linear Algebra , 1997 .

[47]  R. Hambleton,et al.  Handbook of Modern Item Response Theory , 1997 .

[48]  John J. Benedetto,et al.  Applied and numerical harmonic analysis , 1997 .

[49]  T. Minka Expectation-Maximization as lower bound maximization , 1998 .

[50]  Terry M. Peters,et al.  The Fourier Transform in Biomedical Engineering , 1998 .

[51]  G. Reinsel Elements of Multivariate Time Series Analysis, 2nd Edition , 1998 .

[52]  É. Moulines,et al.  Convergence of a stochastic approximation version of the EM algorithm , 1999 .

[53]  Michael E. Tipping,et al.  Probabilistic Principal Component Analysis , 1999 .

[54]  Gianluca Bontempi,et al.  Local learning techniques for modeling, prediction and control , 2000 .

[55]  Carl D. Meyer,et al.  Matrix Analysis and Applied Linear Algebra , 2000 .

[56]  Sigurd Skogestad,et al.  Plantwide controlA review and a new design procedure ” , 2013 .

[57]  N. Levenberg,et al.  Function theory in several complex variables , 2001 .

[58]  Allan P. Donsig,et al.  Real Analysis with Real Applications , 2001 .

[59]  Marco Lippi,et al.  THE GENERALIZED DYNAMIC FACTOR MODEL: REPRESENTATION THEORY , 2001, Econometric Theory.

[60]  D. Brillinger Time series - data analysis and theory , 1981, Classics in applied mathematics.

[61]  T. Chonavel,et al.  Statistical Signal Processing: Modelling and Estimation , 2002 .

[62]  J. Stock,et al.  Macroeconomic Forecasting Using Diffusion Indexes , 2002 .

[63]  Y. Pawitan In all likelihood : statistical modelling and inference using likelihood , 2002 .

[64]  W. D. Haan VECTOR AUTOREGRESSIVE COVARIANCE MATRIX ESTIMATION , 2003 .

[65]  Aapo Hyvärinen,et al.  Independent Component Analysis for Non-Normal Factor Analysis , 2003 .

[66]  T. Lancaster An Introduction to Modern Bayesian Econometrics , 2004 .

[67]  Fabrice Labeau,et al.  Discrete Time Signal Processing , 2004 .

[68]  Andrzej Cichocki,et al.  On-line Algorithm for Blind Signal Extraction of Arbitrarily Distributed, but Temporally Correlated Sources Using Second Order Statistics , 2000, Neural Processing Letters.

[69]  Seungjin Choi Blind Source Separation and Independent Component Analysis : A Review , 2004 .

[70]  B. Ripley,et al.  Robust Statistics , 2018, Encyclopedia of Mathematical Geosciences.

[71]  G Scheler,et al.  Magnetoencephalography (MEG) predicts focal epileptogenicity in cavernomas , 2004, Journal of Neurology, Neurosurgery & Psychiatry.

[72]  S. Kay Intuitive Probability and Random Processes using MATLAB , 2005 .

[73]  Soo-Young Lee Blind Source Separation and Independent Component Analysis: A Review , 2005 .

[74]  Michael J. Panik Advanced Statistics from an Elementary Point of View , 2005 .

[75]  Robert H. Shumway,et al.  Time Series Analysis and Its Applications (Springer Texts in Statistics) , 2005 .

[76]  K. Strimmer,et al.  Statistical Applications in Genetics and Molecular Biology A Shrinkage Approach to Large-Scale Covariance Matrix Estimation and Implications for Functional Genomics , 2011 .

[77]  Robert H. Shumway,et al.  Discrimination and Clustering for Multivariate Time Series , 1998 .

[78]  Charles E. Heckler,et al.  Applied Multivariate Statistical Analysis , 2005, Technometrics.

[79]  Ali Esmaili,et al.  Probability and Random Processes , 2005, Technometrics.

[80]  Aurél Galántai,et al.  Jordan's principal angles in complex vector spaces , 2006, Numer. Linear Algebra Appl..

[81]  Errol C. Caby An Introduction to Statistical Signal Processing , 2006, Technometrics.

[82]  T. Hughes,et al.  Signals and systems , 2006, Genome Biology.

[83]  Kaare Brandt Petersen,et al.  The Matrix Cookbook , 2006 .

[84]  Nathaniel E. Helwig,et al.  An Introduction to Linear Algebra , 2006 .

[85]  Gianluca Bontempi,et al.  New Routes from Minimal Approximation Error to Principal Components , 2008, Neural Processing Letters.

[86]  Manfred K. Warmuth,et al.  Online kernel PCA with entropic matrix updates , 2007, ICML '07.

[87]  Konstantinos Metaxoglou,et al.  Maximum Likelihood Estimation of VARMA Models Using a State‐Space EM Algorithm , 2007 .

[88]  竹安 数博,et al.  Time series analysis and its applications , 2007 .

[89]  Helmut Ltkepohl,et al.  New Introduction to Multiple Time Series Analysis , 2007 .

[90]  Are Hjørungnes,et al.  Complex-Valued Matrix Differentiation: Techniques and Key Results , 2007, IEEE Transactions on Signal Processing.

[91]  V. Bogachev Gaussian Measures on a , 2022 .

[92]  Jim Hefferon,et al.  Linear Algebra , 2012 .

[93]  A. Leon-Garcia,et al.  Probability, statistics, and random processes for electrical engineering , 2008 .

[94]  Alan Julian Izenman,et al.  Modern Multivariate Statistical Techniques: Regression, Classification, and Manifold Learning , 2008 .

[95]  E. El-Darzi,et al.  Analysis of stopping criteria for the EM algorithm in the context of patient grouping according to length of stay , 2008, 2008 4th International IEEE Conference Intelligent Systems.

[96]  P. Deb Finite Mixture Models , 2008 .

[97]  R. Darlington,et al.  Factor Analysis , 2008 .

[98]  Anil K. Jain Data clustering: 50 years beyond K-means , 2008, Pattern Recognit. Lett..

[99]  Leland B. Jackson,et al.  Digital Filters and Signal Processing: With MATLAB® Exercises , 2010 .

[100]  Matti Laine,et al.  Auditory and Cognitive Deficits Associated with Acquired Amusia after Stroke: A Magnetoencephalography and Neuropsychological Follow-Up Study , 2010, PloS one.

[101]  U. Cherubini,et al.  Fourier Transform Methods in Finance , 2010 .

[102]  Philipp Birken,et al.  Numerical Linear Algebra , 2011, Encyclopedia of Parallel Computing.

[103]  David J. Bartholomew,et al.  Latent Variable Models and Factor Analysis: A Unified Approach , 2011 .

[104]  Richard E. Chandler,et al.  Statistical Methods for Trend Detection and Analysis in the Environmental Sciences , 2011 .

[105]  Scott L. Miller,et al.  Probability and Random Processes : With Applications to Signal Processing and Communications , 2012 .

[106]  Seungjin Choi,et al.  Independent Component Analysis , 2009, Handbook of Natural Computing.

[107]  Jean-Paul Chilès,et al.  Wiley Series in Probability and Statistics , 2012 .

[108]  也谈“nowcasting”的汉译 , 2012 .

[109]  V. Neuhaus,et al.  Latent Class Analysis , 2010 .

[110]  J. Wooldridge Introduction to Econometrics , 2013 .