DOCSLIB.ORG

Uncertainty Representation and Interpretation in Model-Based Prognostics Algorithms Based on Kalman Filter Estimation

Total Page：16

File Type：pdf, Size：1020Kb

Download full-text PDF Read full-text
Uncertainty Representation and Interpretation in Model-Based Prognostics Algorithms Based on Kalman Filter Estimation Uncertainty Representation and Interpretation in Model-based Prognostics Algorithms based on Kalman Filter Estimation Jose´ R. Celaya1, Abhinav Saxena2, and Kai Goebel3 1, 2 SGT Inc. NASA Ames Research Center, Moffett Field, CA, 94035, USA [email protected] [email protected] 3 NASA Ames Research Center, Moffett Field, CA, 94035, USA [email protected] ABSTRACT casting of the health parameters is required up to a future time that results in crossing of the pre-established failure condition This article discusses several aspects of uncertainty represen- threshold. This is ultimately required in order to compute re- tation and management for model-based prognostics method- maining useful life. ologies based on our experience with Kalman Filters when applied to prognostics for electronics components. In par- Previous work applied to electrolytic capacitor and power ticular, it explores the implications of modeling remaining MOSFETs (Metal-Oxide Semiconductor Field-Effect Tran- useful life prediction as a stochastic process and how it re- sistor) has focused on implementation of the previously lates to uncertainty representation, management, and the role described process and has presented remaining useful of prognostics in decision-making. A distinction between the life results without any uncertainty measure associated to interpretations of estimated remaining useful life probability them (J. R. Celaya et al., 2011; J. Celaya, Saxena, Kulka- density function and the true remaining useful life probabil- rni, et al., 2012; J. Celaya et al., 2011; J. Celaya, Kulkarni, ity density function is explained and a cautionary argument et al., 2012). Other work on prognostics based on particle fil- is provided against mixing interpretations for the two while tering has been presented regarding remaining useful life as considering prognostics in making critical decisions. a random variable and presenting corresponding uncertainty estimates (Saha et al., 2009; Daigle & Goebel, 2011). This 1. INTRODUCTION work focuses on reviewing uncertainty representation tech- niques used in model-based prognostics and on providing an Model-based prognostics methodologies in electronics prog- interpretation of uncertainty for the electronics prognostics nostics have been developed based on Bayesian tracking applications previously presented, and based on Kalman fil- methods such as Kalman Filter, Extended Kalman Filter, and ter approaches for health state estimation. Particle Filter. The models used in these methodologies are mathematical abstractions of the time evolution of the degra- The Bayesian tracking framework allows for modeling of dation process and the cornerstone for the estimation of re- sources of uncertainty in the measurement process and also maining useful life. The Bayesian tracking framework allows on the degradation evolution dynamic model as applied on for estimation of state of health parameters in prognostics the application under consideration. This is done in terms making use of available measurements from the system under of an additive noise in the model, which is regarded as zero consideration. In this framework, health parameters are re- mean and normally distributed random variable. This allows garded as random variables for which, in the case of Kalman for the aggregation of different sources of uncertainty for the and Extended Kalman filters, their distribution are regarded health state tracking step. Its implications on the uncertainty as Normal and the estimation process focuses on computing estimation for remaining useful life (RUL) including future estimates of the expected value and variance as they relate to state forecasting are discussed in this paper. the mean and variance that fully parametrize the Normal dis- tribution. In addition to the health estimation process, fore- 1.1. Model-based prognostics background Jose´ R. Celaya et al. This is an open-access article distributed under the terms As mentioned earlier, a model-based prognostics methodol- of the Creative Commons Attribution 3.0 United States License, which per- ogy based on Bayesian tracking consists of two steps, health mits unrestricted use, distribution, and reproduction in any medium, provided the original author and source are credited. state estimation and RUL prediction. The following is a high 1 Annual Conference of the Prognostics and Health Management Society 2012 level description of the process that will help to provide the health state estimation step is typically used as initial state appropriate context for the upcoming discussion. value for forecasting x(t) up to tEOL. Remaining useful life R(tp) at time of prediction tp is defined as State of health estimation: To initiate the prediction, it is R(tp) = tEOL tp; (2) − necessary to first establish a starting point, the current state where t is deterministic and known, and t is a random of health. A model-based algorithm employs dynamic mod- p EOL variable function of the failure threshold and the state esti- els of the physical behavior of the system or component under mate x(t ). This function includes the state forecasting step consideration, along with dynamic degradation models of key p and the identification of when the failure threshold is crossed. parameters that represent the degradation over time. Bayesian tracking algorithms like Kalman filter, extended Kalman fil- 1.2. Ideas explored in this paper ter, and particle filter are among the algorithms typically em- ployed in a model-based prognostics methodology (Daigle & In this paper we explore how the state vector variable should Goebel, 2011; Saha & Goebel, 2009; J. Celaya et al., 2011; be interpreted during the tracking phase and how it is related J. Celaya, Saxena, Kulkarni, et al., 2012). In such methodolo- to the process of final RUL prediction. This probability inter- gies, dynamic models of the nominal system and degradation pretation is often overlooked in the literature by interpreting models are posed as a discrete state-space system in which the state vector as the health indicator and a threshold is used the state variable x(t) consists of physical variables, and in on this variable in order to compute EOL (end-of-life) and some cases, it includes degradation model parameters to be RUL. estimated online. Here, we discuss how the state estimation process is defined The models consist of a state equation representing the time in the Bayesian framework. We will, in particular, focus evolution of the state as shown in Eq. (1a); where u(t) is on the output of the estimation process in the Kalman filter the system input and w(t) is a zero-mean and normally dis- framework. Furthermore, we try to interpret the objective tributed additive noise representing random model error. In of the Kalman filter, whether to estimate x(t) as a random addition, the measurement equation (Eq. (1b)) relates the variable or to estimate a parameter of the probability density state variable to measurements of the systems y(t). The term function of x(t) –such as expected value or variance– or both. v(t) is a zero-mean and normally distributed additive noise In addition, we will challenge how we usually think about representing the random measurement error. The measure- RUL and how it has been interpreted using other, similar, ment and model noise normality assumption could be relaxed methods. The main objective here is to characterize its impact when using computational Bayesian methods like particle fil- on uncertainty representation and management. For instance, tering. if RUL is considered as a random variable and we assume that x_ (t) = f(x(t); u(t)) + w(t) (1a) a model-based prognostics framework based on the Kalman filter generates RUL with a particular variance, then it is in- y(t) = h(x(t)); u(t)) + v(t) (1b) correct to arbitrarily expect, assume, or force the variance to be small. The variance of random variables such as RUL is The state of the system, as it evolves through time, is periodi- not under our control as explained in the next section. cally estimated by the filter as measurements y(t) of key vari- ables become available through the life of the system. This These concepts are discussed in the context of prognostics of is the health state estimation step of the model-based prog- electronics, particularly, the uncertainty propagation in power nostics algorithm. Typically, in a model-based prognostics MOSFET and capacitor prognostics applications as presented method, a Bayesian tracking algorithm attempts to estimate in J. R. Celaya et al. (2011); J. Celaya, Saxena, Kulkarni, et the expected value of the joint probability density function of al. (2012) and J. Celaya et al. (2011); J. Celaya, Kulkarni, et the state x(tp). Where tp is the time at which a remaining al. (2012) respectively. In these applications, uncertainty has useful life prediction is computed using only system observa- not been explicitly considered in the prediction results and tions up to this point in time. Different assumptions about the this paper is an effort towards augmenting the methods used probability density function are used depending on the filter there with an uncertainty management methodology. used. 1.3. Background on Uncertainty Management There are several different types of sources of uncertainty that Remaining useful life estimation (prediction): In order to must be accounted for in a prognostic system formulation. compute remaining useful life, the state-equation (Eq. (1a)) These sources may be categorized into following four cat- of the model is used to compute the state evolution in a fore- egories and accordingly require separate representation and casting mode until an end-of-life threshold is reached at time management methods. denoted by tEOL. The last state estimate at time tp in the 2 Annual Conference of the Prognostics and Health Management Society 2012 1. Aleatoric or Statistical Uncertainties: these uncertain- in Engel (2009). Quantification of uncertainty from various ties arise from inherent variability in any process and sources in a process has been investigated and a sensitivity cannot be eliminated.
Load more

Recommended publications
  • Superprocesses and Mckean-Vlasov Equations with Creation of Mass
    Sup erpro cesses and McKean-Vlasov equations with creation of mass L. Overb eck Department of Statistics, University of California, Berkeley, 367, Evans Hall Berkeley, CA 94720, y U.S.A. Abstract Weak solutions of McKean-Vlasov equations with creation of mass are given in terms of sup erpro cesses. The solutions can b e approxi- mated by a sequence of non-interacting sup erpro cesses or by the mean- eld of multityp e sup erpro cesses with mean- eld interaction. The lat- ter approximation is asso ciated with a propagation of chaos statement for weakly interacting multityp e sup erpro cesses. Running title: Sup erpro cesses and McKean-Vlasov equations . 1 Intro duction Sup erpro cesses are useful in solving nonlinear partial di erential equation of 1+ the typ e f = f , 2 0; 1], cf. [Dy]. Wenowchange the p oint of view and showhowtheyprovide sto chastic solutions of nonlinear partial di erential Supp orted byanFellowship of the Deutsche Forschungsgemeinschaft. y On leave from the Universitat Bonn, Institut fur Angewandte Mathematik, Wegelerstr. 6, 53115 Bonn, Germany. 1 equation of McKean-Vlasovtyp e, i.e. wewant to nd weak solutions of d d 2 X X @ @ @ + d x; + bx; : 1.1 = a x; t i t t t t t ij t @t @x @x @x i j i i=1 i;j =1 d Aweak solution = 2 C [0;T];MIR satis es s Z 2 t X X @ @ a f = f + f + d f + b f ds: s ij s t 0 i s s @x @x @x 0 i j i Equation 1.1 generalizes McKean-Vlasov equations of twodi erenttyp es.
    [Show full text]
  • Birth and Death Process in Mean Field Type Interaction
    BIRTH AND DEATH PROCESS IN MEAN FIELD TYPE INTERACTION MARIE-NOÉMIE THAI ABSTRACT. Theaim ofthispaperis to study theasymptoticbehaviorofa system of birth and death processes in mean field type interaction in discrete space. We first establish the exponential convergence of the particle system to equilibrium for a suitable Wasserstein coupling distance. The approach provides an explicit quantitative estimate on the rate of convergence. We prove next a uniform propagation of chaos property. As a consequence, we show that the limit of the associated empirical distribution, which is the solution of a nonlinear differential equation, converges exponentially fast to equilibrium. This paper can be seen as a discrete version of the particle approximation of the McKean-Vlasov equations and is inspired from previous works of Malrieu and al and Caputo, Dai Pra and Posta. AMS 2000 Mathematical Subject Classification: 60K35, 60K25, 60J27, 60B10, 37A25. Keywords: Interacting particle system - mean field - coupling - Wasserstein distance - propagation of chaos. CONTENTS 1. Introduction 1 Long time behavior of the particle system 4 Propagation of chaos 5 Longtimebehaviorofthenonlinearprocess 7 2. Proof of Theorem1.1 7 3. Proof of Theorem1.2 12 4. Proof of Theorem1.5 15 5. Appendix 16 References 18 1. INTRODUCTION The concept of mean field interaction arised in statistical physics with Kac [17] and then arXiv:1510.03238v1 [math.PR] 12 Oct 2015 McKean [21] in order to describe the collisions between particles in a gas, and has later been applied in other areas such as biology or communication networks. A particle system is in mean field interaction when the system acts over one fixed particle through the em- pirical measure of the system.
    [Show full text]
  • 1 Introduction
    The analysis of marked and weighted empirical processes of estimated residuals Vanessa Berenguer-Rico Department of Economics; University of Oxford; Oxford; OX1 3UQ; UK and Mansfield College; Oxford; OX1 3TF [email protected] Søren Johansen Department of Economics; University of Copenhagen and CREATES; Department of Economics and Business; Aarhus University [email protected] Bent Nielsen1 Department of Economics; University of Oxford; Oxford; OX1 3UQ; UK and Nuffield College; Oxford; OX1 1NF; U.K. bent.nielsen@nuffield.ox.ac.uk 29 April 2019 Summary An extended and improved theory is presented for marked and weighted empirical processes of residuals of time series regressions. The theory is motivated by 1- step Huber-skip estimators, where a set of good observations are selected using an initial estimator and an updated estimator is found by applying least squares to the selected observations. In this case, the weights and marks represent powers of the regressors and the regression errors, respectively. The inclusion of marks is a non-trivial extention to previous theory and requires refined martingale arguments. Keywords 1-step Huber-skip; Non-stationarity; Robust Statistics; Stationarity. 1 Introduction We consider marked and weighted empirical processes of residuals from a linear time series regression. Such processes are sums of products of an adapted weight, a mark that is a power of the innovations and an indicator for the residuals belonging to a half line. They have previously been studied by Johansen & Nielsen (2016a) - JN16 henceforth - generalising results by Koul & Ossiander (1994) and Koul (2002) for processes without marks. The results presented extend and improve upon expansions previously given in JN16, while correcting a mistake in the argument, simplifying proofs and allowing weaker conditions on the innovation distribution and regressors.
    [Show full text]
  • Analysis of Error Propagation in Particle Filters with Approximation
    The Annals of Applied Probability 2011, Vol. 21, No. 6, 2343–2378 DOI: 10.1214/11-AAP760 c Institute of Mathematical Statistics, 2011 ANALYSIS OF ERROR PROPAGATION IN PARTICLE FILTERS WITH APPROXIMATION1 By Boris N. Oreshkin and Mark J. Coates McGill University This paper examines the impact of approximation steps that be- come necessary when particle filters are implemented on resource- constrained platforms. We consider particle filters that perform in- termittent approximation, either by subsampling the particles or by generating a parametric approximation. For such algorithms, we de- rive time-uniform bounds on the weak-sense Lp error and present associated exponential inequalities. We motivate the theoretical anal- ysis by considering the leader node particle filter and present numeri- cal experiments exploring its performance and the relationship to the error bounds. 1. Introduction. Particle filters have proven to be an effective approach for addressing difficult tracking problems [8]. Since they are more compu- tationally demanding and require more memory than most other filtering algorithms, they are really only a valid choice for challenging problems, for which other well-established techniques perform poorly. Such problems in- volve dynamics and/or observation models that are substantially nonlinear and non-Gaussian. A particle filter maintains a set of “particles” that are candidate state values of the system (e.g., the position and velocity of an object). The filter evaluates how well individual particles correspond to the dynamic model and set of observations, and updates weights accordingly. The set of weighted particles provides a pointwise approximation to the filtering distribution, which represents the posterior probability of the state.
    [Show full text]
  • The Tail Empirical Process of Regularly Varying Functions of Geometrically Ergodic Markov Chains Rafal Kulik, Philippe Soulier, Olivier Wintenberger
    The tail empirical process of regularly varying functions of geometrically ergodic Markov chains Rafal Kulik, Philippe Soulier, Olivier Wintenberger To cite this version: Rafal Kulik, Philippe Soulier, Olivier Wintenberger. The tail empirical process of regularly vary- ing functions of geometrically ergodic Markov chains. Stochastic Processes and their Applications, Elsevier, 2019, 129 (11), pp.4209-4238. 10.1016/j.spa.2018.11.014. hal-01228825v2 HAL Id: hal-01228825 https://hal.archives-ouvertes.fr/hal-01228825v2 Submitted on 21 Sep 2018 HAL is a multi-disciplinary open access L’archive ouverte pluridisciplinaire HAL, est archive for the deposit and dissemination of sci- destinée au dépôt et à la diffusion de documents entific research documents, whether they are pub- scientifiques de niveau recherche, publiés ou non, lished or not. The documents may come from émanant des établissements d’enseignement et de teaching and research institutions in France or recherche français ou étrangers, des laboratoires abroad, or from public or private research centers. publics ou privés. The tail empirical process of regularly varying functions of geometrically ergodic Markov chains Rafa lKulik∗ Philippe Soulier† Olivier Wintenberger‡ September 21, 2018 Abstract We consider a stationary regularly varying time series which can be expressed as a function of a geometrically ergodic Markov chain. We obtain practical con- ditions for the weak convergence of the tail array sums and feasible estimators of cluster statistics. These conditions include the so-called geometric drift or Foster- Lyapunov condition and can be easily checked for most usual time series models with a Markovian structure. We illustrate these conditions on several models and statis- tical applications.
    [Show full text]
  • Specification Tests for the Error Distribution in GARCH Models
    Computational Statistics and Data Analysis 56 (2012) 3587–3598 Contents lists available at SciVerse ScienceDirect Computational Statistics and Data Analysis journal homepage: www.elsevier.com/locate/csda Specification tests for the error distribution in GARCH models B. Klar a,∗, F. Lindner a, S.G. Meintanis b a Institut für Stochastik, Karlsruhe Institute of Technology (KIT), Kaiserstraße 89, 76133 Karlsruhe, Germany b Department of Economics, National and Kapodistrian University of Athens, 8 Pesmazoglou Street, 105 59 Athens, Greece article info a b s t r a c t Article history: Goodness-of-fit and symmetry tests are proposed for the innovation distribution in Received 19 November 2009 generalized autoregressive conditionally heteroscedastic models. The tests utilize an Received in revised form 18 May 2010 integrated distance involving the empirical characteristic function (or the empirical Laplace Accepted 19 May 2010 transform) computed from properly standardized observations. A bootstrap version of Available online 12 June 2010 the tests serves the purpose of studying the small sample behaviour of the proclaimed procedures in comparison with more classical approaches. Finally, all tests are applied to Keywords: some financial data sets. GARCH model Goodness-of-fit test ' 2010 Elsevier B.V. All rights reserved. Symmetry test Empirical characteristic function Bootstrap test 1. Introduction Suppose that a process is driven by a generalized autoregressive conditionally heteroscedastic (GARCH) model of specific order. This class of models was introduced by Bollerslev(1986), and despite the fact that certain properties of GARCH processes make no use of the particular form of the innovation distribution (see, for example, the stationarity conditions mentioned later in this section), GARCH models with Gaussian innovations have been considered.
    [Show full text]
  • Nonparametric Bayes: Inference Under Nonignorable Missingness and Model Selection
    NONPARAMETRIC BAYES: INFERENCE UNDER NONIGNORABLE MISSINGNESS AND MODEL SELECTION By ANTONIO LINERO A DISSERTATION PRESENTED TO THE GRADUATE SCHOOL OF THE UNIVERSITY OF FLORIDA IN PARTIAL FULFILLMENT OF THE REQUIREMENTS FOR THE DEGREE OF DOCTOR OF PHILOSOPHY UNIVERSITY OF FLORIDA 2015 c 2015 Antonio Linero Dedicated to Mike and Hani for their gracious support, and Katie for her patience! ACKNOWLEDGMENTS I would first like to convey my sincerest gratitude to my advisers, Professor Michael J. Daniels and Professor Hani Doss, for their help and encouragement. I have been fortunate to have two excellent mentors to bounce ideas off of, and to offer assurances when I doubted my work. They have provided me with every opportunity to succeed, and it has been an honor to work with them. I would also like to thank my committee members, Professor Malay Ghosh who always had an open door and Professor Arunava Banerjee for providing a valuable Machine Learning perspective. Outside of my committee, I am grateful to Professor Daniel O. Scharfstein for his insights into the missing data problem and for giving me the opportunity to visit Johns Hopkins. I am also particularly grateful to Professor Ron Randles, whose Introduction to Mathematical Statistics course provided the initial inspiration for me to pursue Statistics, and to Professor Andrew Rosalsky for providing me with a deep appreciation of Probability. Lastly, I would like to thank my partner Katie for her inspiration and patience, and my parents, without whose support I would not have had the opportunity to succeed. 4 TABLE OF CONTENTS page ACKNOWLEDGMENTS.................................4 LIST OF TABLES.....................................8 LIST OF FIGURES....................................9 ABSTRACT.......................................
    [Show full text]
  • Notes on Counting Process in Survival Analysis
    Notes on Counting Process in Survival Analysis Mai Zhou Distribution of some statistics, especially rank based, with censored data are hard to get. Even if they can be worked out for small samples, they are likely to be prohibitively expensive for larger sample sizes. Therefore we need to work on the asymptotic distribution - which often works well for moderate and large sample sizes. This notes is intended to be read along with Fleming and Harrington's book Counting Pro- cesses and Survival Analysis. It is not intended as a rigorous treatment of the subject of counting process martingale. The aim is to (1) present intuitions to help visualize the counting process and (2) supply simplified proofs (in special cases, or with more assumptions, perhaps), make the book easier to read and (3) select the materials that can be realistically covered in one semester or even one quarter. The students need only to have had background in Poisson process and basic probability/statistics. Some prior knowledge of survival analysis will help understand the example/application of counting process in the survival analysis. There is an equally good book on the counting processes: by Andersen, Borgan, Keiding, Gill Statistical Models Based on Counting Processes, Springer 1993 This book contains many more examples of application. The book also has more materials then can be covered in one semester. At a math level similar to FH book. Chapter 5 of Kalbflisch and Prentice, second edition is similar to this note. Review of Master level Survival Analysis. Review of central limit theorem (Linderberg condi- tion) and Poisson process (outlined below).
    [Show full text]
  • Heavy Tailed Analysis Eurandom Summer 2005
    HEAVY TAILED ANALYSIS EURANDOM SUMMER 2005 SIDNEY RESNICK School of Operations Research and Industrial Engineering Cornell University Ithaca, NY 14853 USA [email protected] http://www.orie.cornell.edu/ sid ∼ & Eurandom http://www.eurandom.tue.nl/people/EURANDOM_chair/eurandom_chair.htm 1 60 SIDNEY RESNICK 8. Asymptotic Normality of the Tail Empirical Measure As in Section 7.1, suppose X , j 1 are iid, non-negative random variables with common { j ≥ } distribution F (x), where F¯ RV−α for α > 0. Continue with the notation in (7.1), (7.2), and (7.3). Define the tail empirical∈ process, 1 n n (8.1) W (y)=√k ǫ (y−1/α, ] F¯(b(n/k)y−1/α) n k Xi/b(n/k) ∞ − k i=1 X =√k ν (y−1/α, ] E(ν (y−1/α, ]) . n ∞ − n ∞ Theorem 7. If (7.1), (7.2), and (7.3) hold, then in D[0, ) ∞ W W n ⇒ where W is Brownian motion on [0, ). ∞ Remark 3. Note, because of regular variation, as n , k/n 0, → ∞ → n −α (8.2) Eν (y−1/α, ]= F¯(b(n/k)y−1/α) y−1/α = y. n ∞ k → For applications to such things as the asymptotic normality of the Hill estimator and other estimators derived from the tail empirical measure, we would prefer the centering in (8.1) be y. However, to make this substitution in (8.1) requires knowing or assuming n (8.3) lim √k F¯(b(n/k)y−1/α) y n→∞ k − exists finite. This is one of the origins for the need of second order regular variation.
    [Show full text]
  • Empirical Processes with Applications in Statistics 3
    Empirical Processes with Applications in Statistics 3. Glivenko–Cantelli and Donsker Theorems Johan Segers UCLouvain (Belgium) Conférence Universitaire de Suisse Occidentale Programme Doctoral en Statistique et Probabilités Appliquées Les Diablerets, February 4–5, 2020 1 / 54 Set-up I i.i.d. random variables X1; X2;::: on some probability space Ω taking values in some measurable space (X; A) with common law P P(Xi 2 B) = P(B); 8B 2 A I Family F ⊂ L1(P) of P-integrable functions f : X! R I Empirical and population measures: for f 2 F , 1 Xn P f = f(X ) n n i i=1 Z Pf = f(x) dP(x) = E[f(Xi)] X I Empirical process: p Gn = n (Pn − P) 2 / 54 X e t z a à à I M 3 / 54 Bounded stochastic processes Assume that the following map is bounded almost surely: F! R : f 7! Pnf − Pf Then we can view Pn − P and Gn as (possibly non-measurable) maps Ω ! `1(F ) I Assume that supf2F jf(x)j < 1 for P-almost every x 2 X and supf2F jPfj < 1 I Often, we will assume that F has a P-integrable envelope F : X! [0; 1), i.e., 8x 2 X; 8f 2 F ; jf(x)j 6 F(x) 4 / 54 a q s f a o X z 5 / 54 Uniform law of large numbers: Glivenko–Cantelli Strong law of large numbers. For all f 2 F , Pnf ! Pf a.s.; n ! 1 Trivially, for every f1;:::; fk 2 F , max jPnf − Pfj ! 0 a.s.; n ! 1 j=1;:::;k Can we make this statement uniform in F ? Definition: Glivenko–Cantelli.
    [Show full text]
  • Introduction to Empirical Processes and Semiparametric Inference1
    This is page i Printer: Opaque this Introduction to Empirical Processes and Semiparametric Inference1 Michael R. Kosorok August 2006 1c 2006 SPRINGER SCIENCE+BUSINESS MEDIA, INC. All rights re- served. Permission is granted to print a copy of this preliminary version for non- commercial purposes but not to distribute it in printed or electronic form. The author would appreciate notification of typos and other suggestions for improve- ment. ii This is page iii Printer: Opaque this Preface The goal of this book is to introduce statisticians, and other researchers with a background in mathematical statistics, to empirical processes and semiparametric inference. These powerful research techniques are surpris- ingly useful for studying large sample properties of statistical estimates from realistically complex models as well as for developing new and im- proved approaches to statistical inference. This book is more a textbook than a research monograph, although some new results are presented in later chapters. The level of the book is more introductory than the seminal work of van der Vaart and Wellner (1996). In fact, another purpose of this work is to help readers prepare for the mathematically advanced van der Vaart and Wellner text, as well as for the semiparametric inference work of Bickel, Klaassen, Ritov and Wellner (1997). These two books, along with Pollard (1990) and chapters 19 and 25 of van der Vaart (1998), formulate a very complete and successful elucida- tion of modern empirical process methods. The present book owes much by the way of inspiration, concept, and notation to these previous works. What is perhaps new is the introductory, gradual and unified way this book introduces the reader to the field.
    [Show full text]
  • Bayesian Nonparametric Models
    Bayesian Nonparametric Models Peter Orbanz, Cambridge University Yee Whye Teh, University College London Related keywords: Bayesian Methods, Prior Probabilities, Dirichlet Process, Gaussian Processes. Definition A Bayesian nonparametric model is a Bayesian model on an infinite-dimensional parameter space. The parameter space is typically chosen as the set of all possi- ble solutions for a given learning problem. For example, in a regression problem the parameter space can be the set of continuous functions, and in a density esti- mation problem the space can consist of all densities. A Bayesian nonparametric model uses only a finite subset of the available parameter dimensions to explain a finite sample of observations, with the set of dimensions chosen depending on the sample, such that the effective complexity of the model (as measured by the number of dimensions used) adapts to the data. Classical adaptive problems, such as nonparametric estimation and model selection, can thus be formulated as Bayesian inference problems. Popular examples of Bayesian nonparametric models include Gaussian process regression, in which the correlation structure is refined with growing sample size, and Dirichlet process mixture models for clustering, which adapt the number of clusters to the complexity of the data. Bayesian nonparametric models have recently been applied to a variety of ma- chine learning problems, including regression, classification, clustering, latent variable modeling, sequential modeling, image segmentation, source separation and grammar induction. Motivation and Background Most of machine learning is concerned with learning an appropriate set of pa- rameters within a model class from training data. The meta level problems of determining appropriate model classes are referred to as model selection or model adaptation.
    [Show full text]