| |
Cai, T. Tony Department of Statistics, The
Wharton
School, University of Pennsylvania
A Root-Unroot Transform and Block Thresholding
Approach to
Density Estimation and Poisson Regression
Abstract: Block thresholding methods have demonstrated considerable successes
in wavelet function estimation, especially in Gaussian nonparametric
regression. Block thresholding increases estimation precision by
utilizing information about neighboring wavelet coefficients.
In this talk we will discuss a root-unroot method for density estimation
and Poisson regression. This method turns these problems into a
nonparametric regression problem and then block thresholding technique
is used for estimation. The procedure is easily implementable. We show
that the resulting estimators adaptively achieve the optimal rate of
convergence over a range of Besov Spaces.
Biography: Tony Cai is Associate Professor
of Statistics
at the Wharton School of the University of Pennsylvania.
His research interests include nonparametric
function estimation, wavelet applications, and functional data analysis.
He earned his Ph.D. from Cornell University in 1996 and
is currently Associate Editor for The Annals of Statistics, JASA,
Statistica Sinca and Statistics Surveys. |
Clyde, Merlise Institute
of Statistics and Decision Sciences, Duke University
Bayesian Function Estimation using Continuous Overcomplete
Dictionaries
Abstract: We consider the nonparametric regression problem of estimating an
unknown function based on noisy data. One approach to this estimation
problem is to represent the function in a series expansion using a
linear combination of basis functions. Overcomplete dictionaries
provide a larger, but redundant collection of generating elements than
a basis, however, coefficients in the expansion are no longer unique.
Despite the non-uniqueness, this has the potential to lead to sparser
representations by using fewer non-zero coefficients. Compound
Poisson random fields and their generalization to Levy random fields
are ideally suited for construction of priors on functions using these
overcomplete representations for the general nonparametric regression
problem, and provide a natural limiting generalization of priors for
the finite dimensional version of the regression problem. The price
for the increased flexibility of the overcomplete representation is
the computational challenge of exploring an infinite dimensional
space. While expressions for posterior modes or posterior
distributions of quantities of interest are not available in closed
form, the prior construction using Levy random fields permits
tractable posterior simulation via a reversible jump Markov chain
Monte Carlo algorithm. Efficient computation is possible because
updates based on adding/deleting or updating single dictionary
elements bypass the need to invert large matrices. Furthermore,
because dictionary elements are only computed as needed, memory
requirements scale linearly with the sample size. In comparison with
other methods, the Levy random field priors provide excellent
performance in terms of both mean squared error. We discuss
applications to protein identification and quantification using
MALDI-TOF mass spectroscopy.
Biography: Merlise Clyde is an Associate Professor of Statistics in the Institute
of Statistics and Decision Sciences at Duke University. She received her
PhD degree in 1993 from the School of Statistics at the University of
Minnesota, Her primary research interest is in model selection and
model uncertainty.
|
Donoho, David Statistics
Department, Stanford
How Multiscale Thinking Creates New Challenges for Statisticians |
Dukic, Vanja Biostatistics,
University of Chicago
AIDS Reporting Delay in the US Cities: Analysis of the Centers for Disease
Control (CDC) Data
Abstract: Time
delay between a new AIDS diagnosis and its report to the Centers for
Disease Control (CDC), historically ranging between couple of weeks and
couple of years, presents a significant problem when trying to predict
future AIDS incidence and health care burden. The reporting delay needs to
be correctly estimated and adjusted for in order to avoid potentially
serious downward bias. We examine case reports from 39 large US cities,
received by the CDC as of the end of December 2001 and published in the
APIDS database. We employ Bayesian multi-resolution methodology to
estimate city-specific hazards of reporting delay, adjusting for patient
covariates and within-city correlation. We describe the ranking of the 39
US cities according to their reporting delay distributions based on the
optimal survival curve ranking (OSCR) procedure. We discuss uncertainty in
the reported delay estimates and in the resulting ranking, and present a
graphical approach to visualize this uncertainty.
Biography: Vanja Dukic
got her PhD in Applied Mathematics at Brown University in
2001. Since then she has been at the University of Chicago as an
assistant professor of biostatistics. Her research interests include
Bayesian hierarchical models for epidemics, meta-analysis, and survival
data, as well as model selection and statistical computing. |
Fryzlewicz, Piotr. Department of
Mathematics, University of Bristol
Unbalanced Haar(-Fisz) methodology for function
estimation and
variance stabilisation
Abstract: The Discrete Unbalanced Haar
(DUH) transform is a decomposition of 1D
signals with respect to an orthonormal Haar-like basis where jumps in the
basis vectors do not necessarily occur in the middle of their support. We
introduce a procedure for estimation in Gaussian noise which consists of
three steps: a DUH transform, thresholding of the decomposition
coefficients, and the inverse DUH transform. We show that our estimator is
mean-square consistent with near-optimal rates for a wide range of
functions, uniformly over DUH bases which are not ``too unbalanced".
An important ingredient of our approach is basis selection. We choose each
basis vector so that it best matches the data at a specific scale and
location, where the latter parameters are determined by the ``parent"
basis
vector. Our estimator performs well and is computable in O(n log n)
operations.
Modifications to the above procedure are needed for other types of noise.
We consider Poisson intensity estimation, as well as a multiplicative
regression set-up occurring in e.g. spectrum or volatility estimation. To
account for the heterogeneity of the data, both the "basis
selection" and "thresholding" steps of our algorithm use the so-called Fisz
variance
stabilising transform, whose main idea is to divide a given Unbalanced
Haar
coefficient by an appropriate function of the corresponding ``smooth"
coefficient. With a small modification, the resulting Unbalanced Haar-Fisz
estimation algorithm can also be used to stabilise the variance of
heterogeneous data and bring their distribution closer to normality.
Biography: Piotr Fryzlewicz obtained an M.Sc. in Mathematics
from Wroclaw University
of Technology, Poland (2000) and a Ph.D. in Statistics from the University
of Bristol, UK (2003). In 2003-2005, he was a Chapman Research Fellow at
Imperial College London. He is now a Lecturer in Statistics in the
Department of Mathematics at the University of Bristol.
His research is in the areas of multiscale methods in statistics,
nonparametric regression and time series, with applications to finance,
astronomy and microarray data. His recent work on locally stationary
financial time series models has been supported by a grant from the
Nuffield Foundation. |
Hannig, Jan. Department of Statistics Colorado
State University
Extreme Value Theory for SiZer
Abstract: SiZer is a powerful method
for exploratory data analysis. In this paper
approximation to the distributions underlying the statistical
inference are investigated, and
large improvements are made in the approximation using extreme value
theory.
This results in improved size, and also in an improved global
inference version of SiZer.
The main points are illustrated with real data and simulated examples.
Biography: Jan Hannig received an
MS. in mathematics from Charles University,
Prague in 1996 and Ph.D. in statistics from Michigan State University
in 2000. He is an Assistant Professor in the Department of Statistics
at Colorado State University. His research interests include
stochastic processes and theoretical statistics. |
Hirakawa, Keigo Department of Statistics, Harvard University
Wavelet-Based Image Processing with Missing Data
Abstract: Suppose an image denoising problem is extended to
simultaneously deal with problems with missing or incomplete pixel
values, either because of mechanical designs (e.g. demosaicing) or
because of distortion (e.g. picture impainting). In the context of
wavelet-based image processing, missing or incomplete pixel poses a
difficult problem because wavelet transform takes a linear combination
of image signal, and thus many, or even all of the noisy wavelet
coefficients are unobserved. In this work, a unified framework for
coupling the EM algorithm with the Bayesian hierarchical modeling of
neighboring wavelet coefficients of image signals is presented.
Within this framework, problems with missing pixels or pixel
components, and hence unobservable wavelet coefficients, are handled
simultaneously with denoising. The hyperparameters of the model are
estimated via the marginal likelihood by the EM algorithm, and a part
of the output of its E-step automatically provides optimal estimates,
given the specified Bayesian model, of the noise-free image. This
unified empirical-Bayes based framework, therefore, offers a
statistically principled and extremely flexible approach to a wide
range of pixel estimation problems including image denoising, image
interpolation, demosaicing, or any combinations of them.
Biography: Keigo Hirakawa received B.S.E. in electrical engineering
from Princeton University (Princeton, NJ) in 2000, M.S. and Ph.D in
electrical and computer engineering from Cornell University (Ithaca,
NY) in 2003 and 2005, respectively, where he was a recipient of the
Lockheed Martin Fellowship. Since 2005, he has been a postdoctoral
research associate in the Department of Statistics at Harvard
University while simultaneously pursuing M.M. in jazz performance
(piano) at New England Conservatory (Boston, MA). He has been an
imaging consultant for Hewlett-Packard, Agilent Technologies, NEC Labs
Japan, and Texas Instruments. Hirakawa's research focuses on color
science and on model-based image processing, including image
denoising, interpolation, model parameter estimation, and missing data
problems.
|
Huo, Xiaming.
School of Industrial and Systems Engineering, Georgia
Institute of Technology Multiscale Methodology and Detectability
Abstract: Multiscale
methodology has been proven to be effective in
determining the state-of-the-art boundaries in some detectability
problems. We review some new results in this direction. More
specifically, the existing results give the asymptotic rate of the
boundaries. More precise distributional results can be derived, and
they reveal more accurate properties of the detectability
boundaries. Some of these results will be presented in my talk.
Potential applications are described. Biography: Xiaoming
Huo received the B.S. degree in mathematics from the
University of Science and Technology, China, in 1993, and the M.S.
degree in electrical engineering and the Ph.D. degree in statistics
from Stanford University, Stanford, CA, in 1997 and 1999,
respectively. He is an Associate Professor with the School of
Industrial and Systems Engineering, Georgia Institute of Technology,
Atlanta. His research interests include statistics and multiscale
methodology. Dr. Huo received first prize in the 30th International
Mathematical Olympiad (IMO), which was held in Braunschweig,
Germany. |
Izem, Rima Statistics Department, Harvard
Analysis of Variation in Manifolds
Abstract: Several statistical methods such as principal component analysis
and analysis of variance are often effective in analyzing
variation in high dimensional data when the space of variation is linear.
However, describing variability is much more difficult when the
data varies along nonlinear modes. Simple examples of nonlinear
variation in functional data are horizontal shift of curves of
common shape, frequency change of acoustic signals of common
shape, or lighting change in images of the same object.
This presentation shows novel data depth functions that would extend
data depth concepts to describe variation of multivariate data
when the space of variation is a manifold or the result of
nonlinear variation in the data. We propose new ways of
defining depth in manifolds and they both respect the geometry of
the support of the distribution. We illustrate these new depth measures
for manifolds of constant curvature or known atlas.
Biography: Rima Izem received her PhD in statistics at the University of North
Carolina at Chapel Hill in May 2004. She has spend the last two year as an
assistant professor at the statistics department at Harvard. Her research
interests are in developing statistical methodologies in Functional Data
Analysis, Spatial Statistics, and Nonparametric statistics. She is
particularly interested in applications to Biology and Economics. |
Kolaczyk, Eric D.
Department of Mathematics and Statistics,
Boston University
Multiscale, Multigranular Statistical Image Segmentation
Abstract:
In the image segmentation problem, one seeks to determine and label homogeneous subregions in an image scene, based on pixel-wise
measurements. Motivated by current challenges in the field of remote sensing land cover characterization, we introduce a framework that allows
for adaptive choice of both the spatial resolution of subregions and the categorical granularity of labels. Our framework is based upon a class of
models we call "mixlets," a blend of recursive dyadic partitions and finite mixture models. The first component allows for sparse
representation of spatial structure at multiple resolutions, while the second enables us to capture the varying degrees of mixing of pure
categories that accompany the use of different resolutions. A segmentation is produced in our framework by selecting an optimal mixlet model, through
complexity-penalized maximum likelihood, and summarizing the information in that model with respect to a categorical hierarchy. Both theoretical
and empirical evaluations of the proposed framework are presented. If time allows, we will also comment on current work towards Bayesian and
spatio-temporal extensions.
Biography: Professor Kolaczyk's research focuses on the statistical modeling and analysis of various types of temporal, spatial,
and network data, with a particular emphasis on the development of methods exploiting inherent sparseness. His work has resulted in new methods
for signal and image denoising, tomographic
image reconstruction, disease mapping and clustering, high-level image analysis in land cover
classification, and sampling and monitoring of computer network structure and traffic. Professor Kolaczyk's publications have appeared in the
literatures on statistical theory and methods, engineering, astronomy, geography, and computer science. His work has been supported by
various grants from the Office of Naval Research and the National Science Foundation.
|
Lee,Thomas C. M. Department of
Statistics Colorado State University
Fiducial Curvewise Confidence Intervals for Wavelet Regression
Abstract: We propose a method for constructing curvewise approximate confidence
intervals for wavelet regression. Our method is based on a recent
generalization of fiducial inference studied by Hannig (2006) and is
an alternative to Bayesian based methods. Preliminary simulation
results suggest good frequentist properties of the proposed method.
Joint work with Jan Hannig.
Biography: Thomas Lee received his B.App.Sc. (Math) degree in 1992,
and the B.Sc. (Hons) (Math) degree with University Medal in 1993, all from the
University of Technology, Sydney, Australia. In 1997 he completed a
Ph.D. degree jointly at Macquarie University and CSIRO Mathematical
and Information Sciences, Sydney, Australia. Currently he is an
Associate Professor at the Department of Statistics, Colorado State
University, USA. His research interests include computational
statistics, wavelet analysis, and digital signal and image processing.
|
Meng, Xiao-Li Department of
Statistics, Harvard
A Crash Course in Wavelet Methods
- Short Course, June 11
Biography:
Xiao-Li Meng is Professor and Chairman of the Department of Statistics at Harvard University.
He is also the co-editor of Statistica Sinica. He was the recipient of the 2001 COPSS
Award, and the recipient of the 2003 Distinguished Achievement Award from ICSA.
He was ranked (by Science Watch) among the world top 25 most
cited authors for articles published and cited during 1991-2001 in mathematical sciences. His degrees
include BS (Fudan Mathematics Department, 1982), Master of Science Diploma (Fudan Mathematics Institute,
1986), Master of Art (Harvard Statistics, 1987), and Ph.D. (Harvard Statistics, 1990).
He taught at The University of Chicago from 1991-2001 before joining Harvard University.
He has served on editorial boards for leading statistical journals such as The Annals of
Statistics, Biometrika, Journal of The American Statistical Association, and Bernoulli. He has
served on numerous national and international professional committees, including chairing the
2004 Joint Statistical Meetings. He is also an elected fellow of American Statistical Association
and of Institute of Mathematical Statistics.
His current research interests include wavelet modelling for signal and image data, statistical
issues in astronomy and astrophysics, modelling and imputation for mental health survey data,
Bayesian ranking and mapping, statistical principles and foundational issues, and Markov chain
Monte Carlo, especially perfect sampling.
|
Meyer, Francois Department
of Electrical Engineering, University of Colorado
Charting a functional atlas from an fMRI dataset
Abstract: The main challenge that we intend to address involves the
charting of a
functional atlas from a functional Magnetic Resonance Imaging (fMRI)
dataset. A large number of internal microscopic variables in the brain and
the scanner contribute to the fMRI signal. However, at a macroscopic scale
many of these variables are coupled, and we can assume that the fMRI
signal
can be described by a small number of parameters, in comparison to the
large number of degrees of freedom of the original dataset. We take
advantage of the implicit low dimensionality of the dataset to construct,
in an unsupervised way, a new parametrization of the dataset. The
new
parametrization creates meaningful clusters allowing the separation of the
dataset into: (1) activated voxels, (2) artefactual signals, and (3) a
clutter formed by the background time series. We have conducted
several
experiments with synthetic and in-vivo data that demonstrate the
performance of our approach.
Biography: Francois Meyer is an Associate Professor of Electrical
Engineering at the
University of Colorado at Boulder. His research interests include signal
and image processing and the analysis of biomedical datasets. He received
a
Ph.D. degree in electrical engineering from INRIA, France, in 1993, and
graduated with Honors from Ecole Nationale Superieure d'Informatique et de
Mathematiques Appliquees, Grenoble, in 1987, with a M.S. in computer
science and applied mathematics. |
Nason, GUY Department of Mathematics, University of Bristol
Multiscale adaptive lifting and some applications
Abstract: Lifting generalizes the multiscale method to a
wide range of
interesting
scenarios. This talk describes the `one-coefficient-at-a-time' lifting
method and explains how it can be applied to irregularly spaced data
and functions on networks. We also mention adaptive lifting, selectively
choosing the `best' basis element as the multiscale decomposition
proceeds. We show how multiscale adaptive lifting can be used to improve
prediction of hydrophobic segments along transmembrane proteins and also
in
estimating delay times in a transportation network.
Biography: Guy P Nason is Professor of
Statistics, Department of
Mathematics,
University of Bristol, U.K. His research interests are: multiscale methods
in statistics, non-stationary time series, variance stabilization,
political
science and statistics, statistical methods for defence and security,
statistics on networks, bioinformatics and data and image fusion.
He is the lead author and maintainer of the free WaveThresh package.
He was EPSRC Advanced Research Fellow from 2000-5 and was awarded the
2001 Guy Medal in Bronze by the Royal Statistical Society (RSS) for work
in
wavelet methods in statistics. He was recently Secretary of RSS's
Research Section and is currently a member of RSS Council. He also is
a member of the IMS, ISI, the Bernoulli Society and the IASC. He is a big
fan of Wallace and Gromit and Shrek2.
|
Ombao, Hernando Department of Statistics, University of Illinois at
Urbana-Champaign
Localized Feature Selection for Discrimination and Classification of
Non-Stationary Signals
Abstract: In this talk, we develop an
automatic procedure for selecting
features for discriminating and classifiying in non-stationary signals.
The feature of interest is the spectrum, which is the decomposition of
variance across frequency. In practice, signals are non-stationary,
that is, the distribution of the variance of signals across frequency
changes over time. Thus, for analyzing such signals, we use the SLEX
(smooth localized complex exponentials) library. The SLEX library
consists of many bases; each basis is composed of localized orthogonal
Fourier-like waveforms. The SLEX provides a natural built-in mechanism
for extracting localized spectral features. In our procedure, we use
a training data set that consists of signals whose group memberships are
known. The best basis from the SLEX library is that which minimizes
prediction error. We show via some simulation studies that our method
is able to consistently identify the most discriminant coefficients
and is able to classify signals to the correct groups at a high rate.
Finally, we apply our method to magnetoencephalograms, recorded from a
standard auditory paired-click paradigm, for classifying subjects in the
control and the schizophrenic groups.
Biography: Hernando Ombao is Associate Professor in the Department of Statistics
at the University of Illinois at Urbana-Champaign. He also holds adjunct
appointments in Psychology, Psychiatry and Cognitive Neuroscience. He
received in Ph.D. degree in Biostatistics at the University of Michigan
in 1999. He is actively developing theory and methods for non-stationary
signals and images and is keenly interested in the applications to
neuroscience and seismology. He currently serves as an Associate Editor
for JASA, Theory and Methods. |
Park, Cheolwoo Department of Statistics, University of
Georgia
Multiscale Analysis on Internet Traffic Data
Abstract: It is important to characterize burstiness of
Internet traffic and find the causes for building models
that can mimic real traffic. To achieve this goal,
exploratory analysis tools and statistical tests are needed,
along with new models for aggregated traffic. This talk
introduces statistical tools based on wavelets and SiZer (SIgnificance of ZERo crossings of the derivative). The
intricate fluctuations of Internet traffic are explored in
various respects and lessons on long range dependence and
nonstationarities from real data analyses are summarized.
Biography: Cheolwoo Park obtained his Ph.D. in Statistics from Seoul
National University, Korea, in 2002. He was Postdoc at the
University of North Carolina, Chapel Hill and the
Statistical and Applied Mathematical Sciences Institute, and
Visiting Assistant Professor at the University of Florida.
He is now Assistant Professor in the Department of
Statistics at the University of Georgia.
His research area covers nonparametric function estimation,
Internet traffic data analysis, machine learning, and fMRI
data analysis.
|
Steele, J. Michael Wharton
School, University of Pennsylvannia
Banquet Speech:
The Kolmogorov membrane and Tukey's Statistical Analog
Biography: J.
Michael Steele has served as the C.F. Koo Professor of Statistics since
joining the Wharton School faculty in 1990. Before coming to Wharton, he
taught at the University of British Columbia, Stanford University, and
Princeton University. He received his B.A. in mathematics from Cornell in
1971 and his Ph.D. in mathematics from Stanford in 1975. Steele’s
main area of research is the statistical modeling of asset returns, and he
is the author of Stochastic Calculus
and Financial Applications, a widely used graduate text. Steele’s
most recent book, The Cauchy-Schwarz
Master Class, was published by Cambridge University Press in the
spring of 2004. |
Vannucci, Marina Department of Statistics, Texas A&M University
Bayesian inference for wavelet-based modelling of functional data
Abstract: In this talk I will describe methodologies for Bayesian
modelling of functional data that incorporate feature extraction.
Practical applications will be classification and clustering problems that
involve functional predictors. Wavelet methods will be used for dimension
reduction. Wavelets are orthogonal transformations that allow the
decomposition of a signal into a set of components, each associated to a
particular scale (or resolution). Wavelets have been successfully employed
in various ways in the analysis of functional data. In the practical
contexts of this talk, curves will be transformed to wavelet coefficients
and Bayesian methods will be used to simultaneously estimate group
structures among observations (curves) while identifying discriminating
local features of the curves via the selection of the
corresponding
wavelet coefficients. I will present applications to real data. In
mass
spectrometry, for example, the identification of peaks related to a
specific outcome, i.e. peaks that discriminate samples or that predict a
clinical response, is of interest. Another example will look at
analysing
data from a study involving high-dimensional, high-frequency tidal
volume
traces measured during an induced panic model in normal humans.
Biography: Dr. Vannucci received a Ph.D. in Statistics in 1996 from
the
University of Florence, Italy. In 1998 she joined the Department of
Statistics at Texas A&M University where she is currently a Full
Professor. Her research has focused on the theory and practice of Bayesian
variable selection techniques and on the development of wavelet-based
statistical models and their applications. Her work is often motivated by
real problems that need to be addressed with suitable statistical methods. |
Vidakovic, Brani Department of Biomedical Engineering,
Georgia Institute of
Technology and Emory University
Wavelets in Bioinformatics: Protein and DNA Random Walks and their
Multiscale Analysis
Abstract: In this talk we overview
several applications of wavelets in
biomedical research. The emphasis is placed on scaling measures of
bio data and their statistical use. In particular, we discuss in
more detail so called DNA random walks.
Functional segmentation is one of the many ways to study DNA and
proteins. For example, a genomic sequence under consideration is
partitioned into segments and these are identified as a particular
functional type of DNA (such as coding or regulatory regions) if
some relevant statistical descriptors match with of
experimentally
verified counterparts. DNA and protein random walks and wavelet based
measures of their (multi)fractality are tools for classifying
segments of DNA and proteins to functional types.
In discussing this application we overview results by C-K. Peng and
coauthors, Allan Arneodo and coauthors, Jonas Almeida, and some
others, on simple 1-D DNA random walks generated by PP
(purine/pyramidine) and WS (weak/strong bonds) polarities. Then we
discus multidimensional and marginal protein random walks and their
application in protein classification.
Biography: Brani Vidakovic is Professor
of Statistics at The Wallace H. Coulter
Department of Biomedical Engineering at Georgia Institute of
Technology and Emory University. He has BS and MS in Mathematics
from University of Belgrade (1978, 1981) and PhD in Statistics from
Purdue University (1992). He was an Assistant and Associate
Professor of Statistics and Decision Sciences at Duke University
prior to joining Georgia Institute of Technology in 2000. Dr
Vidakovic is currently the President of Georgia Chapter of American
Statistical Association. He is an Editor-in-Chief of Wiley's
Encyclopedia of Statistical Sciences and Associate Editor of several
statistical journals. His research interests include wavelets, Bayesian
and
computational statistics, nonparametrics, and biostatistics. |
von Sachs, Rainer Institut de
statistique, Université catholique de Louvain
A multiscale approach for statistical
characterization of functional brain images
Abstract: In this talk we present an approach
of spatial multiscales for an
improved characterization of functional pixel intensities of
(medical) images. Examples are numerous such as temporal dependence of
brain response intensities measured by fMRI or frequency dependence of NMR
spectra measured at each pixel. The overall goal is to improve the
misclassification rate in (unsupervised) clustering of the functional
image
content into a finite but unknown number of classes. Hereby we adopt a
non-parametric point of view to reduce the functional dimensionality of
the
observed pixel intensities, by statistical aggregation of
non-linear wavelet threshold estimators of these intensity curves. As we
model them by a very general functional form, this is opposed to
commonly
used parametric feature extraction based on a priori knowledge on the
nature of the functional response.
The same paradigm applies to our spatial multiscale approach used to
improve upon the low degree of information aggregation of monoscale
statistical models used to extract structure in the underlying noisy
measurements on the pixel scale. Instead of modelling correlation between
neighboring pixels we use an approach based on Recursive Dyadic
Partitioning of the image. Complexity-penalised maximum likelihood
estimation based on Gaussian mixture models (in the domain of the discrete
wavelet transform of the pixel intensity curves), which we estimate via EM,
will allow us to choose the locally best scales for clustering the image
content into a finite but adaptively chosen number of cluster classes.
In this talk we present results both on the theoretical treatment of the
encountered estimation steps and on the numerical performance of our
algorithm on simulated and real data examples.
This is joint work with Anestis Antoniadis (Université Joseph Fourier,
Grenoble) and Jérémie Bigot (Université Paul Sabbatier, Toulouse).
Biography: Rainer von
Sachs obtained his Ph.D. in Mathematics 1991 from the University
of Heidelberg, Germany. Until 1998 he was with the Mathematics Department
of the University of Kaiserslautern, Germany. Since then he is Professor
of
Statistics at the Institut de statistique, Université catholique de
Louvain, Louvain-la-Neuve, Belgium. His main research interests are
nonparametric curve estimation, time series analysis, wavelets and
multiscale methods. Rainer von Sachs is an elected member of ISI and
member
of the Bernoulli Society and the IMS. He currently serves as Associate
Editor of the Journal of the Royal Statistical Society, Series B.
|
Wang,Yazhen Statistics,
University of Connecticut
Multiscale Jump and Volatility Analysis for
High-Frequency Financial Data
Abstract: Volatilities of asset returns are pivotal for many
issues in
financial economics. The availability of high frequency intraday
data should allow us to estimate volatility more accurately.
Asset prices often contain jumps, and high-frequency financial data
are inevitably contaminated with market microstructure noise.
Existing methods can deal with noisy data for the continuous
diffusion price model or handle the jump-diffusion price model
without noise. This talk will present estimation of integrated
volatility and jump variation for noisy high-frequency financial
data with jumps. The proposed wavelet based multi-scale methodology
can cope with both jumps in the price and market microstructure
noise in the data, and estimate both integrated volatility and jump
variation from the noisy data. We establish convergence rates for
the proposed estimators of integrated volatility and jump variation.
In particular, we show that the integrated volatility can be
estimated asymptotically under the jump-diffusion price model as
well as under the continuous diffusion price model. Simulations are
conducted to assess the performance of the proposed estimators and
to compare them with existing ones. Theoretical and numerical
analysis show that the proposed estimators outperform existing
methods for noisy high-frequency data under the jump-diffusion
model, and have comparable performance for the continuous diffusion
model and noiseless jump-diffusion model. The methods are illustrated
by applications to two high-frequency exchange rate data sets.
This talk is based on a joint work with Jianqing Fan.
Biography
Yazhen Wang is Professor of Statistics at the University of Connecticut.
He obtained his Ph.D in statistics from University of California at
Berkeley in 1992. His research interests are financial econometrics,
nonparametric curve estimation, change points, long-memory process and
self-similar process, wavelets and multiscale methods, and order
restricted statistical inference. Yazhen Wang is an IMS fellow and an
elected member of ISI. He currently serves as an Associate Editor of
Statistica Sinica. |
Wolf, Patrick Statistics
Department, Harvard
A Crash Course in Wavelet Methods
- Short Course, June 11
Abstract: Patrick J. Wolfe received a B.S. in Electrical Engineering and a B.Mus.
concurrently from the University of Illinois <http://www.uiuc.edu>
at
Urbana-Champaign, both with honors. He earned his Ph.D. in Engineering
from the University of Cambridge <http://www.cam.ac.uk>
as a US National
Science Foundation Graduate Research Fellow, working on the application
of perceptual criteria to statistical audio signal processing.
Prior to joining Harvard in 2004, Professor Wolfe held a Fellowship and
College Lectureship jointly in Engineering and Computer Science at New
Hall <http://www.newhall.cam.ac.uk>,
a Cambridge College where he also
served as Dean. He has also taught in the Department of Statistical
Science <http://www.ucl.ac.uk/Stats/>
at University College, London, and
continues to act as a consultant to the professional audio community.
In addition to his diverse teaching activities, Professor Wolfe has
published in the literatures of engineering, computer science, and
statistics, and has received honors from the Acoustical Society of
America and the International Society for Bayesian Analysis. His
research interests lie at the intersection of statistical signal
processing and numerical harmonic analysis, and encompass general as
well as audio-related applications.Fwd shortcourse urgent |
|