You are Visitor No:
|
Note: the program is not final and is subject to possible changes
First Term
| 4, November | Haim Ricas, Tashtit Scientific Consultants |
|
"Modern statistical software tools.
I. Modern applied statistics with S-Plus"
| |
| 18, November | Haim Ricas, Tashtit Scientific Consultants |
|
"Modern statistical software tools.
II. Statistical software for teaching statistics"
| |
| 2, December | Camil Fuchs, Tel Aviv University |
|
"Lifetime morbid risks in family studies: classical methods and a new model"
| |
| 9, December | A Special Day of Seminars on Robust Design |
| Ron Kenett, KPA Ltd. | |
| "Robust design and rapid development from computer simulations" | |
| Hila Ginsburg, Tel Aviv University | |
|
"Designing experiments in robust design problems"
| |
| 30, December | Yechezkel (Eli) Kling, Tel Aviv University |
|
"Aspects of multiplicity in statistical process control"
| |
| 20, January | Stephen Fienberg, Carnegie Mellon University |
| "Characterizing probability distributions associated with multi-dimensional contingency tables" |
Second Term
| 2, March | Ulrich Stadtmuller, University of Ulm |
|
"Generalized linear models with functional data"
| |
| 16, March | Laurence Freedman, Bar Ilan University |
|
"A new method for dealing with measurement error in
explanatory variables of regression models"
| |
| 23, March | Dan Geiger, Technion |
| "A new software for genetic linkage analysis"
| |
| 30, March | Vyacheslav Abramov, Tel Aviv University |
| "Asymptotic methods for communication networks"
| |
| 20, April | Daniel Yekutieli, Tel Aviv University |
| "FDR confidence intervals"
| |
| 4, May | Albert Vexler, Central Bureau of Statistics |
| "Guarenteed maximum likelihood splitting tests
of a linear regression model"
| |
| 18, May | Yoav Dvir, Tel Aviv University |
| "Local likelihood methods for nonparametric density estimation"
| |
| 1, June | Inna Stainvas, Orbotech Ltd. |
|
"A generative probabilistic oriented wavelet model for texture segmentation"
| |
| 8, June | Yuval Nov, Stanford University |
|
"Modeling and analysis of protein design under resource constraints"
|
Seminars are held on Tuesdays, 10.30 am, Schreiber Building, 309 (see the TAU map ).
is served before.
The seminar organiser is Felix Abramovich . To join the seminar mailing list and get updated information about current/forthcoming seminars and for other inquiries call (03)-6405389 or email felix@math.tau.ac.il
Details of previous seminars:
ABSTRACTS
Haim Ricas
"Modern statistical software tools.
I. Modern applied statistics with S-Plus"
In this talk we present the new S-Plus modules designed for various problems
in modern applied statistics :
- SpatialStats (Spatial Data Analysis)
- EnvironmentalStats (Environmental Data Analysis)
- Seqtrial (Clinical Trials)
- FinMetrics (Modern and flexible analytics for powerful econometric analysis)
- Insightful Miner (Data mining for Massive Data Sets Using Visual Programming)
- S+ ArrayAnalyzer (Statistical Methodology for Microarray analysis)
Haim Ricas
"Modern statistical software tools.
II. Statistical software for teaching statistics"
In this talk we demonstrate how the new statistical software can be
helpful in teaching statistics. We present MathStatica -
the new Mathematica module that allows both symbolic and numerical
solutions to various probability and statistical problems. In particular,
we illustrate how MathStatica can assist in teaching probability,
estimation, statistical inference and asymptotic theory.
We also introduce TableCurve - a new software for curve fitting and
equation discovery which can be an efficient tool for exploratory
parametric and non-parametric data fitting .
Family studies assess routinely the lifetime morbid risks of various diseases
of the relatives of probands affected either by the studied diseases or by other
related diseases.
In family studies in the psychiatric literature the lifetime morbid risks are usually determined either by methods originally designed for analyzing life tables, as the Kaplan-Meier product limit estimator and Cox proportional hazards model, or by simpler estimators like the Weinberg abridged method, or by the Stromgren method which can be considered as an elaboration to the Weinberg abridged method. In other cases, the lifetime morbid risks are assessed by the unadjusted proportion of the affected in the sample (known as the lifetime prevalence).
We shall show that the use of the Kaplan-Meier product limit estimator for the
estimation of lifetime morbid risk may yield unreliable estimates. Furthermore,
while the simplicity of the Stromgren method and the Weinberg abridged method
is appealing, we suggest that under a proper model, those methods can be
replaced by an equally simple statistic, which is shown to be a more accurate
both on the average as well in the great majority of the specific cases. The
increased accuracy is achieved particularly when the investigators do have some
prior indification on the distribution of the ages at onset for those affected
by the disorder.
Modern companies are under increasing pressure to reduce development
time and to provide robust products. The TITOSIM project, funded by
the European Community and headed by Fiat Research, developed statistical
methodology and tools for using computer simulations to achieve these
product development goals. Computer experiments are often conducted in
order to optimize product performance while respecting constraints that
may be imposed. Several methods for achieving robust design in this
context are described and compared with the aid of a simple example
problem. The methods presented compare classical as well as modern
approaches and introduce the idea of a "stochastic response" to aid
the search for robust solutions. Emphasis is placed on the efficiency
of each method with respect to computational cost and the ability to
formulate objectives that encapsulate the notion of robustness.
This is joint work with Ron Bates and David Steinberg.
In the last decades the method of Robust Design that was originally
suggested by Taguchi has been widely applied to various engineering
areas. Usually, when a designer aims for a robust design of a system
with unknown analytical form, he or she follows a two-step procedure.
First, he or she fits a response function for the unknown system by
using experimental arrays that are based on known design of experiments
(DOE) criteria. Second, once the response function has been established,
he or she formulates a Robust-Design criterion and solves it to obtain
an optimal robust configuration.
In this work, we aim to combine both steps in a unified yet sequential DOE protocol. In particular, we suggest a methodology for designing experiments that minimize the variance of the optimal robust configuration. In other words, the variance of the optimal solution for the robust system is minimized already at the DOE stage. This new DOE optimal criteria prioritizes the various response's coefficients and enables the designer to indicate which coefficients should be estimated more accurately with respect to others in order to obtain a reliable robust solution.
The suggested method provides more information on the optimal robust
solution by generating a (multidimensional) distribution of it. Numerical
examples will be presented for illustration purposes.
The area of Statistical Process Control (SPC) is explored in search of
practical uses for the FDR. First, the "Multiplicity Problem", is reviewed
in the context of Statistical Process Control (SPC). A few SPC situations
that give rise to multiplicity are discussed and the concept of the p-value
in this context is examined. Several possibilities for the incorporation of
the p-values into SPC graphical display are examined. The p-valued SPC chart
suggested is a simple, consistent, and intuitive display. This type of
presentation enables the utilization of multiplicity protection methods
without inhibiting the lay-users. The appropriateness of the FWE and FDR in
the SPC situations is reviewed. Furthermore, a new family measure is
constructed based on the discovery costs - the False Discovery Cost Ratio
(FDCR). A p-value base FDCR controlling procedure is obtained by applying
the Benjamini-Hochberg (1997) Weighted Linear Step-up procedure to the set
of p-values corresponding to the individual hypotheses and to the
intersection hypothesis; weighting them by the appropriate discovery costs.
Three possible statistics for testing the intersection hypothesis are
examined: Simes' statistic, Fisher's statistic and Hotelling's T2. It is
shown that when the Simes' statistic is used for this purpose it ensures
that the FDCR is controlled by the proposed procedure.
We review alternative ways to characterize probability distributions associated
with two-way tables of counts (contingency tables) using marginals,
conditionals, and odds ratios and their generalizations to higher dimensions.
Partial specification of such distributions arises in a number of statistical
contexts and usually involves the use of log-linear models or the dropping of
components from complete specifications. We link both of these approaches to
recent developments in algebraic geometry and discuss the insights and new
tools that such linkages bring to statistical methodology. Practical statistical
problems arising in disclosure limitation have provided ongoing motivation to
these developments as well as an outlet for application.
In the talk a generalized linear regression model for a regression situation
where the response variable is a scalar and the predictor is a random function
will be proposed. A linear predictor is obtained by forming the scalar product
of the predictor function with a smooth parameter function, and the expected
value of the response is related to this linear predictor via a link function.
If, in addition, a variance function is specified, this leads to a functional
estimating equation which corresponds to maximizing a functional
quasi-likelihood. This general approach includes the special cases of the
functional linear model, as well as functional Poisson regression and functional
binomial regression. The latter leads to procedures for classification and
discrimination of stochastic processes and functional data. We also consider
the situation where link and variance functions are unknown and are estimated
nonparametrically from the data. As an application, the classification of
medflies in regard to their remaining longevity status depending on their
fertility will be discussed.
I will introduce a new method, moment reconstruction, of correcting for
measurement error in covariates in regression models. The central idea is
similar to regression calibration in that the values of the covariates that
are measured with error are replaced by "adjusted" values. In regression
calibration the adjusted value is the expectation of the true value
conditional on the measured value. In moment reconstruction the adjusted
value is a variance-preserving shrinkage estimate of the true value
conditional on the outcome variable. The adjusted values have the same first
two moments and the same covariance with the outcome variable as the
unobserved "true" covariate values. Unlike regression calibration, moment
reconstruction can deal with differential measurement error. For
case-control studies with logistic regression and covariates that are
normally distributed within cases and controls, the resulting estimates of
the regression coefficients are consistent. In simulations of logistic
regression, moment reconstruction carries less bias than regression
calibration, and for case-control studies is superior in mean square error
to the standard regression calibration approach.
I will give an example of the use of moment reconstruction in linear
discriminant analysis and a non-standard problem where we wish to
adjust a classification tree for measurement error in the explanatory variables.
Genetic linkage analysis is a useful statistical tool for mapping disease
genes and for associating functionality of genes to their location on the
chromosome. I will describe a program, called SUPERLINK, for linkage
analysis and demonstrate its performance. I will focus on the relevant
combinatorial problems that need to be solved in order to optimize the
running time and space requirements of these types of programs, and on some
new capabilities of this software. The talk is intended for audience with
no background in Genetics.
Joint work with Ma'ayan Fishelson.
This talk is concerned with the study of non-Markovian queueing systems and
networks, with applications to communication networks. Its main contribution
consists in deriving results for non-Markovian systems that have been obtained
so far only for Markovian queueing systems. We study large closed client/server
communication networks and losses in single-server queueing systems, with an
application to communication networks of loss queues. We apply stochastic
calculus and the theory of martingales to the case when one of the client
stations is a bottleneck in the limit, where the total number of tasks in the
server increases to infinity. The main results of this study are (i) an
explicit expression for the interrelation between the limiting non-stationary
distributions in non-bottleneck client stations; thus when one distribution is
found in a simulation, the others can be computed; (ii) derivation of diffusion
and fluid approximations for the non-Markovian queue length in the bottleneck
client station. For the loss networks considered, we find an asymptotic
expression for the loss probability and other performance measures, as buffer
capacity increases to infinity. We also find the changes in the loss probability
when redundant packets are added to the messages. The application of martingale
methods for the study of the asymptotic behavior of non-Markovian queueing
systems seems to be new.
Confidence intervals are often constructed only for parameters selected after
viewing the data. The problem with this practice is that the selected intervals
fail to provide the assumed coverage probability. To overcome this problem I
will introduce the FCR - a measure of the intervals` coverage following
selection, and a general procedure offering FCR control under any selection
rule. I will discuss the new procedure and its relation to the Benjamini-Hochberg
procedure and bring theoretical results for independent and positively dependent
test statistics.
We propose and examine a class of generalized maximum likelihood asymptotic
power one tests for detection of various types of changes in a linear
regression model. In economic and epidemiologic studies, such segmented
regression models often occur as threshold models, where it is assumed that
the exposure has no influence on the response up to a possible unknown
threshold. An important task of such studies is testing the existence and
estimation of this threshold. Guaranteed non-asymptotic upper bounds for the
significance levels of these tests are presented. We demonstrate how the
proposed tests were applied toward solving an actual problem encountered
with real data.
An application: According to one theoretical hypothesis, the revenue of
establishments that engage in Research and Development in Israeli economy
and have a small number of employed persons does not depend on the
technological intensity of establishment's activities (e.g. Griliches and
Regev (1999)). This observation might be associated with the fact that the
criteria for categorizing a "small" firm are not clearly defined. However,
if we do not reject the aforementioned hypothesis, it is possible to
estimate threshold number of employed persons that define a firm as "small".
To investigate this question, we used the proposed approach presented in
this talk.
Local likelihood density estimation methods have the advantage that they
provide high order bias reduction for multivariate density estimation.
We review different methods of local likelihood and present analysis of
the asymptotic behavior for the different methods. Then we propose two
new estimators that rely on the current local likelihood methods. This
work is done under the supervision of Prof. David Steinberg.
This talk addresses image segmentation via a generative model approach.
A Bayesian network (BNT) in the space of dyadic wavelet transform
coefficients is introduced to model texture images. The model is similar to a
Hidden Markov model (HMM), but with non-stationary transitive conditional
probability distributions. It is composed of discrete hidden variables and
observable Gaussian outputs for wavelet coefficients. In particular, the Gabor
wavelet transform is considered. The introduced model is compared with the
simplest joint Gaussian probabilistic model for Gabor wavelet coefficients for
several textures from the Brodatz album. The comparison is based on
cross-validation and includes probabilistic model ensembles instead of single
models. In addition, the robustness of the models to cope with additive Gaussian
noise is investigated. We further study the feasibility of the introduced
generative model for image segmentation in the novelty detection framework. Two
examples are considered: (i) sea surface pollution detection from intensity
images and (ii) image segmentation of the still images with varying illumination
across the scene.
Joint work with Prof. Larry Wein.