|
|
|
Colloquia
|
| December 4, 2009, 10:10 am |
Dr. Ian Dryden: Department of Statistics of University of South Carolina |
| November 20, 2009, 10:10 am |
Jelani Wiltshire |
| November 13, 2009, 10:10 am |
Dr. Jun Liu: Department of Statistics, Harvard University |
| November 6, 2009, 10:10 am |
Dr. Ted Chang: Department of Statistics, University of Virginia |
| October 30, 2009, 10:10 am |
Dr. Bala Rajaratnam: Department of Statistics, Stanford University |
| October 26, 2009, 3:40 am |
Muffasir Badshah: Ph.D. Candidate, Department of Statistics, FSU |
| October 23, 2009, 10:10 am |
Haiyan Zhao: Ph.D. Candidate, Department of Statistics, FSU |
| October 9, 2009, 10:10 am |
Dr. Hamid Krim: Electrical and Computer Engineering Department, NCSU |
| October 2, 2009, 10:10 am |
Dr. Anuj Srivastava |
| September 25, 2009, 10:10 am |
Adrian Peter: Northrop Grumman |
| September 16, 2009, 10:00 am |
Sutan Wu: Ph.D. Candidate, Department of Statistics, FSU |
| September 11, 2009, 10:10 am |
Greg Miller: Ph.D Candidate, Department of Statistics, FSU |
| May 1, 2009, 1:30 pm |
Shuva Gupta |
| April 30, 2009, 2:00 pm |
Wenhao Gui |
| April 24, 2009, 10:10 am |
Vernon Lawhern and Dr. Wei Wu, Department of Statistics, FSU |
| April 22, 2009, 3:30 pm |
Michael Crane |
| April 10, 2009, 10:10 am |
Dr. Alan Yuille, Department of Statistics, UCLA |
| April 8, 2009, 9:45 am |
Lanjia Lin |
| April 3, 2009, 10:10 am |
Dr. Michael Black, Department of Computer Science, Brown University |
| March 27, 2009, 10:10 am |
Dr. David Gilbert, Department of Biological Science, FSU |
| March 25, 2009, 11:05 am |
Nikolay Balov |
| March 23, 2009, 3:30 pm |
Yang Liu |
| March 20, 2009, 10:10 am |
Dr. Frits Ruymgaart, Department of Mathematics and Statistics, Texas Tech University |
| March 16, 2009, 3:30 pm |
Dr. Robert Kass, Department of Statistics, Carnegie Mellon University |
| March 6, 2009, 10:10 am |
Dr. Yiyuan She, Department of Statistics, FSU |
| February 27, 2009, 10:10 am |
Dr. Yiyuan She, Department of Statistics, FSU |
| February 13, 2009, 10:10 am |
Dr. Jinfeng Zhang, Department of Statistics, FSU |
| February 6, 2009, 10:10 am |
Dr. Adrian Barbu, Department of Statistics, FSU |
| December 4, 2009 |
| Dr. Ian Dryden: Department of Statistics of University of South Carolina |
| Bayesian Alignment of Unlabeled Marked Point Sets Using Random Fields |
| December 4, 2009 10:10 am |
| 110 OSB |
|
| In structural bioinformatics and chemoinformatics
it is of great interest to align molecules, but the
task is often very difficult. Statistical methodology
is proposed for comparing unlabeled marked point sets,
with an application to aligning steroid molecules in
chemoinformatics. Methods from statistical shape
analysis are combined with techniques for predicting
random fields in spatial statistics in order to
define a suitable measure of similarity between two molecules.
Bayesian modeling of the predicted field overlap between
pairs of molecules is proposed, and posterior inference
of the alignment is carried out using Markov chain Monte
Carlo simulation. By representing the fields in
reproducing kernel Hilbert spaces, the
degree of molecule overlap can be computed without
expensive numerical integration. Superimposing
entire fields rather than the configuration matrices of
point co--ordinates thereby avoids the problem that there
is usually no clear one--to--one correspondence between
the atoms. Using a similar concept, we also propose an
adaptation of the generalized Procrustes analysis algorithm
for the simultaneous alignment of multiple point sets. The
methodology is illustrated with a simulation study and then
applied to the dataset of 31 steroid molecules, where the
relationship between shape and binding activity to the
corticosteroid binding globulin receptor is explored.
This is joint work with Irina Czogiel and Chris Brignell.
|
| Back To Top |
| November 20, 2009 |
| Jelani Wiltshire |
| Age effects in the extinction of Planktonic Foraminifera: A new look at Van Valen's Red Queen hypothesis |
| November 20, 2009 10:10 am |
| 110 OSB |
|
| Van Valen's Red Queen hypothesis stated that within a homogeneous taxonomic group the age is statistically independent of the rate of extinction. The case of the Red Queen hypothesis being addressed here is when the homogeneous taxonomic group is a group of similar species. Since Van Valen's work, various statistical approaches have been used to address the relationship between taxon duration (age) and the rate of extinction. Some of the more recent approaches to this problem using Planktonic Foraminifera (Foram) extinction data include Weibull and Exponential modeling (Parker and Arnold, 1997), and Cox proportional hazards modeling (Doran, 2004,2006). I propose a general class of test statistics that can be used to test for the effect of age on extinction. These test statistics allow for a varying background rate of extinction and attempt to remove the effects of other covariates when assessing the effect of age on extinction. No model is assumed for the covariate effects. Instead I control for covariate effects by pairing or grouping together similar species. In my presentation I will apply my test statistics to the Foram data and to simulated data sets. |
| Back To Top |
| November 13, 2009 |
| Dr. Jun Liu: Department of Statistics, Harvard University |
| Bayesian Partition Models for Detecting Interactions |
| November 13, 2009 10:10 am |
| 110 OSB |
|
| Suppose we have N individuals and for each individual we observed its response vector variable (Yi1,…, Yiq) and its p-dimensional categorical-valued covariates (Xi1,…, Xip). Our goal is to discover which subset of the response variables is influenced by which subset of the covariates. Although the problem is similar to the multiple-response regression, our goal is much more ambitious than just finding certain linear relationships. I will present a novel Bayesian partition model through the use of a set of latent indicator vectors, together with a Markov chain Monte Carlo algorithm, to tackle the problem. I will illustrate the power of the method mainly using examples in genome-wide genetic association studies and in studies of expression quantitative trait loci (eQTL).
Because of the large number of genes and genetic markers in such studies, it is extremely challenging to discover how a small number of genetic markers interact with each other to affect the expression levels for a set of genes (or values of a set of other traits of interest). Our extensive simulation studies mimicked such data structures and demonstrated that our Bayesian approach is much more powerful than the standard two-stage step-wise approach as practiced by statisticians and geneticists in solving “e-QTL” problems.
|
| Back To Top |
| November 6, 2009 |
| Dr. Ted Chang: Department of Statistics, University of Virginia |
| Residual Processes for Joint Motion |
| November 6, 2009 10:10 am |
| 110 OSB |
|
| Cut and paste this link to get the full text. http://stat.fsu.edu/colloquium/Changabstract.pdf |
| Back To Top |
| October 30, 2009 |
| Dr. Bala Rajaratnam: Department of Statistics, Stanford University |
| Flexible Covariance Estimation in Gaussian Graphical models |
| October 30, 2009 10:10 am |
| 110 OSB |
|
| Covariance estimation is known to be a challenging problem, especially for high-dimensional data. In this context, graphical models can act as a tool for regularization and have proven to be excellent tools for the analysis of high dimensional data. Graphical models are statistical models where dependencies between variables are represented by means of a graph. Both frequentist and Bayesian inferential procedures for graphical models have recently received much attention in the statistics literature. The hyper-inverse Wishart distribution is a commonly used prior for Bayesian inference on covariance matrices in Gaussian Graphical models. This prior has the distinct advantage that it is a conjugate prior for this model but it suffers from lack of flexibility in high dimensional problems due to its single shape parameter. In this talk, for posterior inference on covariance matrices in decomposable Gaussian graphical models, we use a flexible class of conjugate prior distributions defined on the cone of positive-definite matrices with fixed zeros according to a graph G. This class includes the hyper inverse Wishart distribution and allows for up to k+1 shape parameters where k denotes the number of cliques in the graph. We first add to this class of priors, a reference prior, which can be viewed as an improper member of this class. We then derive the general form of the Bayes estimators under traditional loss functions adapted to graphical models and exploit the conjugacy relationship in these models to express these estimators in closed form. The closed form solutions allow us to avoid heavy computational costs that are usually incurred in these high-dimensional problems. We also investigate decision-theoretic properties of the standard frequentist estimator, which is the maximum likelihood estimator, in these problems. Furthermore, we illustrate the performance of our estimators by exploring frequentist risk properties and the efficacy of graphs in the estimation of high-dimensional covariance structures. We demonstrate that our estimators yield substantial risk reductions over the maximum likelihood estimator in the graphical model. |
| Back To Top |
| October 26, 2009 |
| Muffasir Badshah: Ph.D. Candidate, Department of Statistics, FSU |
| Computation of the Wealth Distribution in a Heterogeneous Agent Economy |
| October 26, 2009 3:40 am |
| 108 OSB |
|
| In 1928 a young mathematician named Frank Ramsey proposed a dynamic model to answer a simple yet difficult question: “How much of its income should a nation save?" Ramsey's model lays a foundation for macroeconomic theory, and variants of his dynamic optimization problem are the cornerstones of most models of economic fluctuations and growth. The present paper reviews recent research aimed at solving a theoretical model of the macroeconomy (economy in a broad sense) with five key elements (i) it is based on rational decision-making by consumers and a single firm; (ii) it is dynamic, so that consumption and savings decisions are determined by intertemporal (current decisions that take into account the future choices) decisions; (iii) it has stochastic aggregate shocks (random uncertainty at economic level) which lead to upswings and downswings at a macroeconomic level; (iv) it considers general equilibrium, so that interest rates and wage rates are determined endogenously- determined by the interaction of entities in the given economy; and (v) it has a heterogeneous population structure where consumers differ in wealth and employment status against which they cannot insure.
In a heterogeneous-agent based economy, wealth is unevenly distributed among consumers, and part of the model solution is to determine this wealth distribution. This helps economists answer questions such as “What would be the effect of changing wage rate (or unemployment benefits) on individual wealth?"... etc. Algorithms to solve heterogeneous agent models with endogenous wealth distributions have been introduced in economic literature in the past 15 years. We present a formal description of the general model we would like to solve using an alternative (and potentially faster) technique that involves computing the wealth distribution using homogeneous and inhomogeneous Markov chains. |
| Back To Top |
| October 23, 2009 |
| Haiyan Zhao: Ph.D. Candidate, Department of Statistics, FSU |
| Time-varying Coefficient Models with ARMA-GARCH Structures for Longitudinal Data Analysis |
| October 23, 2009 10:10 am |
| 110 OSB |
|
| The motivation of my research comes from the analysis of Framingham Heart Study (FHS) data. FHS is a long term prospective study of cardiovascular disease in the community of Framingham, Massachusetts. The study began in 1948 and 5,209 subjects were initially enrolled in the study. Examinations were given biennially to the study participants and their status associated with occurrence of disease was recorded. The event we are interested in is the occurrence of coronary heart disease (CHD). Covariates considered in this study include gender, age, smoking status, serum cholesterol (SCL), systolic blood pressure (SBP) and body mass index (BMI, weight in kilograms/height in meter squared). Statistical literature indicated that the effects of covariates on CHD may change with time. For example, low cholesterol may decrease CHD risk under age 50 years. However, there is no increased CHD risk with either high or low cholesterol for patients older than 50. So the erect of cholesterol level changes over time. To capture this feature, time-varying coefficient models with ARMA-GARCH structure are developed in this dissertation research. Maximum likelihood and marginal
likelihood methods are used to estimate the parameters in the proposed models. Since high-dimensional integrals are involved in the calculations of marginal likelihood, the Laplace approximation is employed in this study. Simulation studies are conducted to evaluate the performance of these two estimation methods for our proposed models. The Kullback-Leibler (KL) divergence is employed in the simulation to compare the performance. The simulation results show that the marginal likelihood approach gives more accurate parameter estimates, but is more computationally intensive. |
| Back To Top |
| October 9, 2009 |
| Dr. Hamid Krim: Electrical and Computer Engineering Department, NCSU |
| Activity Classification by Shape modeling |
| October 9, 2009 10:10 am |
| 110 OSB |
|
| In this talk, we describe our effort on shape based human activity modeling. From video data, we extract shape sequences to represent the recorded activities. Each of these sequences is actually a curve on a shape manifold. Curves from the same class of activity are realizations of an underlying random process. By lifing realization to a Euclidean Space, we realize our process on a flat space and proceed with the classification.. We use a moving frame and a parallel transport that is intrinsic to the manifold. We also note that our work builds on the shape manifold first proposed by Klas. Srivastava and Mio. We illustrate the models and demonstrate them on some specific activities.
Bio:
Dr. Hamid Krim, Director, Vision Information and Statistical Signal Theories and Applications (VISSTA) and professor in the Department of Electrical and Computer Engineering at North Carolina State University. His research interests are in Statistical Signal and Image Analysis and Mathematical Modeling.
|
| Back To Top |
| October 2, 2009 |
| Dr. Anuj Srivastava |
| A Theory for Statistical Analysis of Shapes of Curves and Surfaces |
| October 2, 2009 10:10 am |
| 110 OSB |
|
| In this talk I will summarize the research of our group in
shape analysis over the past five years, and will present
some future directions. The main goal in shape analysis
is to match, deform, and compare any two objects in such
a way that rotation, translation, and scaling any one of them
does not change the result. Furthermore, in case of parameterized
objects, the analysis should also be invariant to parameterizations.
We have developed a Riemannian framework to accomplish
this task -- the shapes are represented by equivalence classes in
Riemannian Hilbert submanifold and are compared by computing
geodesic paths in the sets of such equivalence classes.
This framework can be used for: (1) quantifying differences between shapes of
two objects, (2) optimally deform one shape into another, (3) computing
sample mean and sample covariance of a set of objects, (4) modeling
variability in a shape class using a wrapped Gaussian (or mixtures of
Gaussians) model on the shape space, (5) performing likelihood ratio
test to classify a novel shape in pre-determined shape classes, (6) quantifying
the level of symmetry of a any given shape, (7) tracking shape variations over
time, and (8) predicting 2D shapes of 3D objects from novel viewing angles.
These tools have been applied to problems in biometrics, military target
recognition, medical image analysis, and bioinformatics, and I will present
some examples of these applications. |
| Back To Top |
| September 25, 2009 |
| Adrian Peter: Northrop Grumman |
| Information Geometry for Shape Analysis: Probabilistic Models for Shape Matching |
| September 25, 2009 10:10 am |
| 110 OSB |
|
| The study of shape analysis is a core field in computer vision. It is a fundamental building block of higher level cognitive tasks such as recognition. This talk will discuss novel approaches to basic shape analysis tasks, including shape matching and defining metrics for shape similarity. Our investigations into these methods will also highlight supporting statistical tools that have general applicability outside the realm of shape analysis, particularly a new wavelet density estimation procedure. All of the techniques are theoretically grounded in the framework of information geometry.
Information geometry is an emerging math discipline that applies differential geometry to space of probability distributions. Within this context, our basic approach to shape analysis is straightforward: represent shapes as probability densities, then use the intrinsic geometry of the space of densities to establish geodesics between shapes. Valid intermediate densities (shapes) can be obtained by walking along the geodesics and the length of the geodesic immediately gives us a similarity measure between shapes. These concepts are illustrated by using two models to represent the probability densities: Gaussian mixtures and wavelet expansions. Beyond shape analysis, information geometry has applications to image processing, data fusion, document mining and a host of areas that require distance/divergence measures between probability distributions.
|
| Back To Top |
| September 16, 2009 |
| Sutan Wu: Ph.D. Candidate, Department of Statistics, FSU |
| Goodness-of-Fit Tests for Logistic Regression and Cox Regression |
| September 16, 2009 10:00 am |
| 307 HCB-Classroom Building |
|
| Goodness-of-fit tests are tools to describe how well a model fits the observations. There are many different goodness-of-fit tests in literature, and we are interested in four of them: Chi-square test, Hosmer-Lemeshow test, Unweighted sum of square test and Cumulative sum of residuals test. These tests are major methods in checking the fitted logistic regression and generalized linear model. However they all have their individual advantages and disadvantages in practical clinical studies. Then which one is the best to describe a particular case? My research will focus on comparing the performances of these methods in practical data and simulation.
Moreover, I am also interested in investigating the interconnection between stacked logistic regression and Cox regression. Cox regression is the most popular model in survival analysis. Diagnostics for Cox model depend on the plots of various kinds of residuals, which may not provide interpretable test results for the model fitting. By adapting the four testing methods to Cox regression, we could get some insight of how well the Cox regression fits the data. |
| Back To Top |
| September 11, 2009 |
| Greg Miller: Ph.D Candidate, Department of Statistics, FSU |
| TBA |
| September 11, 2009 10:10 am |
| 110 OSB |
|
| Risk assessment models are tools that are used to help determine a
patient's risk for treatment purposes. Two widely used tools for
determining risk of heart disease are developed from Framingham Study and
the SCORE project. The risk assessment tool based on Framingham study,
developed by Wilson et al., uses Coronary Heart Disease(CHD) morbidity as
an endpoint. The risk assessment tool from SCORE project is based on 12
European cohort studies where Cardiovascular Disease (CVD) mortality is
used as an endpoint. This raises the primary question: Would different
treatment decisions be made depending on whether one examines risk of
mortality or risk of morbidity? The purpose of this essay is to
investigate the primary question by examining related concepts such as
surrogate variables and validation. Data from the Framingham study is
used to calculate two different Cox proportional hazards models, and
bootstrap sampling is used to compare coefficients from the two models.
The discussion ends with comments about future work and goals. |
| Back To Top |
| May 1, 2009 |
| Shuva Gupta |
| Dissertation Defense - A Study of the Asymptotic Properties of Lasso Estimates for Correlated Data |
| May 1, 2009 1:30 pm |
| OSB 210E, Conference Room |
|
| In this thesis we investigate post-model selection properties of L1 penalized weighted
least squares estimators in regression models with a large number of variables M and
correlated errors. We focus on correct subset selection and on the asymptotic distribution of the penalized estimators. In the simple case of AR(1) errors we give conditions under which correct subset selection can be achieved via our procedure. We then provide a detailed generalization of this result to models with errors that have a weak-dependency structure (Doukhan 1996). We also explore various special cases of weak dependence and show what the theorem translates to in these special cases. We then use estimates of the concentration matrix in the minimization criterion and investigate the results when the LASSO estimates are calculated from the altered criterion function. In all cases, the number M of regression variables is allowed to exceed the sample size n. We further investigate the asymptotic distribution of our estimates, when M < n, and show that under appropriate choices of the tuning parameters the limiting
distribution is multivariate normal. This generalizes to the case of correlated errors the
result of Knight and Fu (2000), obtained for regression models with independent errors. |
| Back To Top |
| April 30, 2009 |
| Wenhao Gui |
| Dissertation Defense - Adaptive series estimators for copula densities |
| April 30, 2009 2:00 pm |
| OSB 210E, Conference Room |
|
| In this thesis, based on an orthonormal series expansion, we propose a new
nonparametric method to estimate copula density functions. We propose
estimators of the variance of the estimated basis coefficients and
establish their consistency. We derive the asymptotic distribution of the
estimated coefficients under mild conditions on the copula density. Based
on this result, we derive an oracle inequality for the copula density
estimator. We propose a stop rule for selecting the number of coefficients
used in the series and we prove that this rule minimizes the mean
integrated squared error. In addition, we consider hard and soft
thresholding techniques for sparse representations. We obtain oracle
inequalities that hold with prescribed probability for various norms of
the difference between the copula density and our estimator. A simulation
study indicates that our method is extremely easy to implement and works
very well, and it compares favorably to the popular kernel based copula
density estimator, especially around the boundary points, in terms of mean
squared error. Finally, we have applied our method to an insurance
dataset. We reach the same conclusion as the parametric methods in the
literature and as such we provide additional justification for the use of
the developed parametric model. |
| Back To Top |
| April 24, 2009 |
| Vernon Lawhern and Dr. Wei Wu, Department of Statistics, FSU |
| Motor cortical decoding using improved generalized linear models |
| April 24, 2009 10:10 am |
| 110 OSB |
|
| Classical generalized linear models (GLMs) have been developed for modeling and decoding neural activity in the motor cortex. These models are based on various forms of Poisson or non-Poisson processes, and provide a reasonable characterization between neural activity and motor behavior. However, they lack a description of other movement-related terms, such as joint angles at the shoulder and elbow, muscular activation, and the subject's attention span. We propose to represent these important, yet unknown states into the GLM model as one multi-dimensional hidden state in the spiking rate density function. The system state in this new model includes two parts: one is the observable behavioral state such as the hand motion, and the other is the hidden state. We tested this new method in two experimental datasets. The results show that the new method significantly improves the model-fitting over the classical GLM model. Moreover, it provides more accurate decoding, while keeping real-time efficiency. These results suggest that this method could contribute as a useful tool to neural prosthetics. |
| Back To Top |
| April 22, 2009 |
| Michael Crane |
| Essay Defense - Nonparametric Estimation of 3D Projective Shapes |
| April 22, 2009 3:30 pm |
| 110 OSB |
|
| This talk is about analysis of invariants of a 3D configuration from its 2D
images in pictures of this configuration, without requiring any
restriction for the camera positioning relative to the scene pictured. We
mention some of the main results found in the literature. The methodology
used is nonparametric, manifold based combined with standard computer
vision reconstruction techniques. More specifically, we use asymptotic
results for the extrinsic sample mean and the extrinsic sample covariance
to construct bootstrap confidence regions for mean projective shapes of 3D
configurations. |
| Back To Top |
| April 10, 2009 |
| Dr. Alan Yuille, Department of Statistics, UCLA |
| Recursive Compositional Models |
| April 10, 2009 10:10 am |
| 110 OSB |
|
| Recursive Compositional Models (RCMs) are a class of probability models designed to detect, recognize, parse, and segment visual objects and label visual scenes. They take into account the statistical and computational complexities of visual patterns. The key design principle is recursive compositionality. Visual patterns are represented by RCMs in a hierarchical form where complex structures are composed of more elementary structures. Probabilities are defined over these structures exploiting properties of the hierarchy (e.g. long range spatial relationships can be represented by local potentials). The compositional nature of this representation enables efficient learning and inference algorithms. Hence the overall architecture of RCMs provides a balance between statistical and computational complexity. HOST: Dr. Adrian Barbu |
| Back To Top |
| April 8, 2009 |
| Lanjia Lin |
| Dissertation Defense - Association Models for Clustered Data with Binary and Continuous Responses |
| April 8, 2009 9:45 am |
| 499 Dirac Science Library Seminar Room |
|
| This dissertation develops novel single random effect models as well as bivariate correlated random effects model for clustered data with mixed bivariate responses. Logit and identity link functions are used for the binary and continuous responses. For the ease of interpretation of the regression effects, random effect of the binary response has bridge distribution so that the marginal model of mean of the binary response after integrating out the random effect preserves the logistic form. And the marginal regression function of the continuous response preserves the linear form. Within-cluster and within-subjects associations could be measured by our proposed models. Fully parametric and semi-parametric Bayesian methods as well as maximum likelihood method are illustrated for model analysis. In the semiparametric Bayesian model, normality assumption of the regression error for the continuous response is relaxed by using a nonparametric Dirichlet Process prior. Robustness of the correlated bivariate random effects model using ML method to misspecifications of regression function as well as random effect distribution is investigated by simulation studies. The Bayesian and likelihood methods are applied to a developmental toxicity study of ethylene glycol in mice. |
| Back To Top |
| April 3, 2009 |
| Dr. Michael Black, Department of Computer Science, Brown University |
| Predicting Human Body Shape Under Clothing |
| April 3, 2009 10:10 am |
| 110 OSB |
|
| We propose a method to estimate the detailed 3D shape of a person from images of that person wearing clothing. The approach exploits a model of human body shapes that is learned from a database of over 2000 range scans. We show that the parameters of this shape model can be recovered independently of body pose. We further propose a generalization of the visual hull to account for the fact that observed silhouettes of clothed people do not provide a tight bound on the true 3D shape. With clothed subjects, different poses provide different constraints on the possible underlying 3D body shape. We consequently combine constraints across pose to more accurately estimate 3D body shape in the presence of occluding clothing. Finally we use the recovered 3D shape to estimate the gender of subjects and then employ gender-specific body models to refine our shape estimates. Results on a novel database of thousands of images of clothed and "naked" subjects, as well as sequences from the HumanEva dataset, suggest the method may be accurate enough for biometric shape analysis in video.
This is joint work with Alexandru Balan.
Project page: http://www.cs.brown.edu/~alb/scapeClothing/
Related ECCV paper: http://www.cs.brown.edu/~black/Papers/balanECCV08.pdf.
HOST: Dr. Wei Wu |
| Back To Top |
| March 27, 2009 |
| Dr. David Gilbert, Department of Biological Science, FSU |
| Epigenomics, Mathematics and Stem Cells |
| March 27, 2009 10:10 am |
| 110 OSB |
|
| I will begin with an informal discussion of how modern technical
advances have transformed biomedical research from the study of one
gene or one protein at a time to the study of all genes and proteins
in each and every experiment. These huge data sets hold great promise
but are now the rate-limiting step for the mathematically challenged
molecular biologist. I will then describe one of these great promises
in a bit more detail- the ability to transform any cell type into any
other cell type to regenerate diseased or deteriorating tissue-and how
our research on chromosome structure and replication relates to this
problem. HOST: Dr. Jinfeng Zhang |
| Back To Top |
| March 25, 2009 |
| Nikolay Balov |
| Dissertation Defense - Covariance on Manifolds |
| March 25, 2009 11:05 am |
| HCB 315 |
|
| With ever increasing complexity of observational and theoretical data models, the sufficiency of the classical statistical techniques, designed to be applied only on vector quantities, is challenged.
Non-linear statistical analysis is an area of intensive research in recent years.
Despite of the impressive progress in this direction, however, a unified and consistent framework is still not present.
In this regard, the following work is an attempt to improve our understanding of random phenomena on non-Euclidean spaces.
More specifically, the motivating goal of the present dissertation is
to generalize the notion of distribution covariance,
which in standard settings is defined only in Euclidean spaces,
on manifolds with metric also known as Riemannian manifolds.
I introduce a tensor field structure, named covariance field,
that is consistent with the heterogeneous nature of manifolds
and not only describes the variability imposed by a distribution,
but in general provides alternative distribution representations.
The covariance field of a distribution combines distribution's density with
geometric characteristics of its domain and thus fills the gap between these two.
I present some of the properties of the covariance fields and argue that
they can be successfully applied to various statistical problems.
In particular, I provide a unified approach for defining parametric families of distributions on manifolds,
parameter estimation using regression analysis,
non-parametric statistical tests for comparing distributions and
interpolation between distributions.
Emphasized are several application areas where this new theory may have potential impact.
One of them is the branch of directional statistics,
with domain of influence ranging from geosciences to medical image analysis.
The fundamental level at which the covariance based structures are introduced,
opens new area for future research. |
| Back To Top |
| March 23, 2009 |
| Yang Liu |
| Dissertation Defense - Transformation Models for Survival Data Analysis and Applications |
| March 23, 2009 3:30 pm |
| 110 OSB |
|
| It is often assumed that all uncensored subjects will eventually experience the event of interest in standard survival models. However, in some situations when the considering event is not death, it will never occur for a proportion of subjects. Survival models with a cure fraction are becoming popular in analyzing this type of studies. We propose a generalized transformation model motivated by Zeng et al's (2006) transformed proportional time cure model. In our proposed model, fractional polynomials are used instead of the simple linear combination of the covariates. The proposed models give us more flexibility without loosing any good properties of the original model, such as asymptotical consistency and asymptotical normality of the regression coefficients. The proposed model will better fit the data where the relationship between a response variable and covariates is non-linear. We also provide a power selection procedure based on likelihood function. A simulation study is curried out to show the accuracy of the proposed power selection procedure. The proposed models are applied to coronary heart disease and cancer related medical data from both observational cohort studies and clinical trials. |
| Back To Top |
| March 20, 2009 |
| Dr. Frits Ruymgaart, Department of Mathematics and Statistics, Texas Tech University |
| Deconvolution and the ruin problem for unknown claims distributions |
| March 20, 2009 10:10 am |
| 214 Duxbury Hall (Nursing) |
|
| The well-known insurance ruin problem is reconsidered. The ruin
probability is estimated in the case of an unknown claims density,
assuming a sample of claims is given. An important step in the
construction of the estimator is the application of a regularized version
of the inverse of the Laplace transform. A rate of convergence in
probability for the ISE is derived and some simulations are included. HOST: Dr. Victor Patrangenaru. |
| Back To Top |
| March 16, 2009 |
| Dr. Robert Kass, Department of Statistics, Carnegie Mellon University |
| Analysis of Neural Spike Train Data: Methods and Commentary |
| March 16, 2009 3:30 pm |
| 110 OSB |
|
| One of the most important techniques in learning about the functioning
of the brain has involved examining neuronal activity in laboratory
animals under varying experimental conditions. Neural information is
represented and communicated through series of action potentials, or
spike trains, and the central scientific issue in many studies
concerns the physiological significance that should be attached to a
particular neuron firing pattern in a particular part of the brain. In
addition, a major comparatively new effort in neurophysiology involves
the use of multielectrode recording, in which responses from dozens of
neurons are recorded simultaneously. Among other things, this has
made possible the construction of brain-controlled robotic devices,
which could benefit people whose movement has been severely impaired.
In my talk I will briefly outline the progress made, by many people,
over the past 10 years, highlighting some of the work my colleagues
and I have contributed. Part of my perspective comes from
the evolution of the international workshops "Statistical Analysis
of Neural Data", which I co-organized in 2002, 2004, 2006, and 2008.
I will try to emphasize general statistical ideas, but will also
indicate current status and future challenges. HOST: Dr. Wei Wu |
| Back To Top |
| March 6, 2009 |
| Dr. Yiyuan She, Department of Statistics, FSU |
| Thresholding-based Iterative Selection Procedures for Model Selection and Shrinkage |
| March 6, 2009 10:10 am |
| 110 OSB |
|
| A class of thresholding-based iterative selection procedures (TISP) is proposed for model selection and shrinkage. People have long before noticed the weakness of the convex $l_1$-constraint (or the soft-thresholding) in wavelets and have designed many different forms of nonconvex penalties to increase model sparsity and accuracy. But for a nonorthogonal regression matrix, there is great difficulty in both investigating the performance in theory and solving the problem in computation. Starting from the thresholding rules rather than penalty functions, TISP provides a simple and efficient way to tackle this.
Moreover, a novel Hybrid-TISP is proposed based on hard-thresholding and ridge-thresholding. It provides a fusion between the $l_0$-penalty and the $l_2$-penalty, and adaptively achieves the right balance between shrinkage and selection in statistical modeling. |
| Back To Top |
| February 27, 2009 |
| Dr. Yiyuan She, Department of Statistics, FSU |
| Outlier Detection via Penalized Regressions |
| February 27, 2009 10:10 am |
| 110 OSB |
|
| Outlier identification has been a classical topic in robust analysis and numerous procedures have been proposed in the past 30 years which are often ad-hoc and computationally expensive. Assuming a mean shift outlier model, we claim that outlier detection amounts to solving a large-p sparsity recovery problem. However, the naive $l_1$-penalty cannot completely remove the masking and swamping effects in the presence of multiple outliers. We build a connection between the penalized mean shift model and Huber's M-estimation functions and develop different versions of iterative procedures to tackle the problem. |
| Back To Top |
| February 13, 2009 |
| Dr. Jinfeng Zhang, Department of Statistics, FSU |
| Automatic Information Extraction from Scientific Literature |
| February 13, 2009 10:10 am |
| 214 Duxbury Hall (Nursing) |
|
| Protein-protein interaction (PPI) extraction from pub-lished
biological articles has attracted much attention because of the
importance of protein interactions in biological processes. Despite
significant progress, mining PPIs from literatures still rely heavily
on time and resource consuming manual annotations. In this study, we
developed a novel methodology based on Bayesian networks (BNs) for
extracting PPI triplets (a PPI triplet consists of two protein names
and the corresponding interaction word) from unstructured text. The
method achieved an overall accuracy of 87% on a cross-validation test
using manually annotated data set. We also showed, through extracting
PPI triplets from a large number of PubMed abstracts, that our method
was able to complement human annotations to extract large number of
new PPIs from literature. Our method performed better in extracting
PPI triplets than, PIE, the best publicly accessible method. |
| Back To Top |
| February 6, 2009 |
| Dr. Adrian Barbu, Department of Statistics, FSU |
| Training an Active Random Field for Real-Time Image Denoising |
| February 6, 2009 10:10 am |
| 214 Duxbury Hall (Nursing) |
|
| Many computer vision problems can be formulated in a Bayesian framework with
Markov Random Field (MRF) or Conditional Random Field (CRF) priors. Given an image, the solution is usually a MAP estimation of the hidden variables of
interest. Recently however, Wainwright observed that in computationally
restricted settings, a model designed for MAP estimation would not give the
best performance and certain biased models might perform better. This talk
will pursued this idea for training real-time image denoising algorithms
using Bayesian models based on MRFs. An Active Random Field (ARF) is defined
as a combination of the MRF/CRF based model and a fast algorithm for hidden
variable estimation. For image denoising, the ARF was chosen to be the
Fields of Experts MRF model and the algorithm was a 1-4 iteration gradient
descent. Finding the ARF parameters is an optimization of a loss function
defined on a training set consisting of pairs of input images and desired
outputs. Experimental validation on unseen data shows that the Active Random
Field approach obtains an improved benchmark performance as well as a
1000-3000 times speedup compared to the Fields of Experts MRF that was
trained for MAP estimation. Using the ARF approach, image denoising can be
performed in real-time, at 8fps on a single CPU for a 256x256 image
sequence, with close to state-of-the-art accuracy. |
| Back To Top |
|
|
|
