Statistics Seminar at Georgia State University
Fall 2013-Spring 2014, Fridays 3:00-4:00pm, Paul Erdos Conference room (796)
Organizer: Yichuan Zhao
If you would like to give a talk in Statistics Seminar, please send an email to
Yichuan Zhao at
2:00-3:00pm, March 14, 796 COE
Professor Jiajia Zhang,
Department of Biostatistics, University of South Carolina
Kernel Smoothed Profile Likelihood Estimation in the Accelerated Failure Time Frailty Model for Clustered Survival Data
Clustered survival data frequently arise in biomedical applications,
where event times of interest are clustered into groups such as families.
In this article we consider an accelerated failure time frailty model
for clustered survival data and develop nonparametric maximum likelihood
estimation for it via a kernel smoother aided EM algorithm. We show that
the proposed estimator for the regression coefficients is consistent, asymptotically
normal and semiparametric efficient when the kernel bandwidth is properly chosen.
An EM-aided numerical differentiation method is derived for estimating its variance.
Simulation studies evaluate the finite sample performance of the estimator, and
it is applied to the Diabetic Retinopathy data set.
3:00-4:00pm, January 16, 796 COE
Dr. Hongqi Xue,
Department of Biostatistics, University of Rochester
Variable Selection for High-Dimensional Nonparametric ODE Models With Applications to Dynamic Gene Regulatory Networks
The gene regulation network (GRN) is a high-dimensional complex system, which can be represented by various mathematical or statistical models. The ordinary differential equation (ODE) model is one of the popular dynamic GRN models. High-dimensional linear ODE models have been proposed to identify GRNs, but with a limitation of the linear regulation effect assumption. We propose a nonparametric additive ODE model, coupled with two-stage smoothing-based ODE estimation methods and adaptive group LASSO techniques, to model dynamic GRNs that could flexibly deal with nonlinear regulation effects. The asymptotic properties of the proposed method are established and simulation studies are performed to validate the proposed approach. An application example for identifying the nonlinear dynamic GRN of T-cell activation is used to illustrate the usefulness of the proposed method.
3:00-4:00pm, January 15, 796 COE
Dr. Bruce Swihart,
Department of Biostatistics, Johns Hopkins University
Sleep Hypnograms, Insurance Claims, & Hand Movement After Stroke: Big Data and Potentially Many Weak, Predictive Signals
The increasing ease and decreasing cost for collecting complex and structured data has ushered in the Big Data era. Three big data sets are discussed in terms of their size and modeling approaches, with a primary focus on sleep hypnogram data. 5,598 sleep hypnograms collected from the Sleep Heart Health Study are analyzed in the context of sleep-disordered breathing and fragmented sleep with scalable GEE log-linear models as well as Multi-state Survival models. The information loss between 5-state hypnograms and 3-state hypnograms is explored. In the second dataset of Insurance Claims, the task of predicting the Days in Hospital of the following year is undertaken as issued in the challenge of the $3 million Heritage Health Prize. Hand Movements After Stroke (Stroke Kinematics) are highly structured and complex data that contain features that are a strong classifier of Stroke-status.
3:00-4:00pm, Januray 14, 796 COE
Professor Jing Zhang,
Department of Statistics, Yale University
Inferring Functional Interaction and Transition Patterns via
Dynamic Bayesian Variable Partition Model
Static pair-wise functional connectivity has been widely used in the neuroimaging field. In contrast, higher-order functional interactions among brain networks and their temporal dynamic transition patterns have been rarely explored. Our recent paper proposes a novel dynamic Bayesian variable partition model (DBVPM) that simultaneously considers and models high-order functional interactions and their dynamics via a unified Bayesian framework. Then, we modeled and characterized these temporal state transitions as finite-state machines, and quantitatively compared their transition patterns between post-traumatic stress disorder (PTSD) patients and healthy controls. We found that these interaction patterns are hopping among a finite number of states, and PTSD patients have a different functional interaction state-space and their temporal transition patterns are substantially different in comparison with healthy controls. This work discovered interesting phenomena that cannot be revealed by static pair-wise functional connectivity, thus offering novel opportunities for deciphering the working mechanisms of brain networks in the future.
3:00-4:00pm, November 22, 796 COE
Professor Xiaojing Ye,
Department of Mathematics and Statistics, Georgia State University
Influence Prediction for Social Networks
We consider the modeling and computations of random dynamical
processes of viral signals propagating over time in social networks.
The viral signals of interests can be popular tweets on trendy topics
in social media such as Twitter or Weibo, or computer malware on the
Internet, or infectious diseases spreading between human or animal hosts.
The viral signal propagations can be modeled as diffusion processes with various
dynamical properties on graphs or networks, which are essentially
different from the classical diffusions carried out in continuous
spaces. The influence is quantified as the expected number of infected
or involved individuals during the viral signal propagation, and hence
is critical in decision-makings in governmental and commercial activities.
We address a critical computational problem in predicting
influences of such signal propagations, and develop a framework using
discrete Fokker-Planck equation method to solve this problem in an
efficient and effective manner.
2:00-3:00pm, October 25, 796 COE
Dr. Zhigang Zhang,
Department of Epidemiology and Biostatistics, Memorial Sloan-Kettering Cancer Center, New York
A class of transformed mean residual life models under right censoring
The mean residual life function is an alternative to the survival function or the hazard function of
a survival time in practice. It provides the remaining life expectancy of a subject surviving up to t. In this study,
we propose a class of transformed mean
residual life models for fitting survival data under right censoring. To estimate the model parameter,
we make use of the inverse probability of censoring weighting approach and develop a system of estimating equations.
Both asymptotic and finite sample properties of the proposed estimators are established and the approach is applied to
two real life data sets collected from a clinical trial. We also considered the efficiency and double robustness of our
estimator, and developed a
model checking technique for one of the special cases of the transformation models.
3:00-4:00pm, October 11, 796 COE
Professor Ajay Subramanian,
Georgia State University
Dynamic Prudential Regulation
We investigate the design of prudential bank regulation in a continuous-time structural
framework. In our model, the regulator controls the dynamic risk-shifting
incentives of a representative bank through the threat of intervention in the bank's
operations. The optimal regulatory policy, which we characterize analytically, entails
an optimal combination of a capital requirement, operational intervention to
control the risk of the bank's portfolio, capital injection, recapitalization, and liquidation
of the bank. The regulator optimally intervenes when the bank's capital ratio
lies inside a "band" consisting of two triggers. We calibrate the model using the simulated
method of moments. The optimal capital requirement exceeds 20%, which
supports the substantially higher capital requirements being proposed in the Basel
III accords. Although capital injection and operational intervention independently
have modest impacts, their optimal combination significantly improves banks' private
and social values. Optimal capital requirements should be counter-cyclical and
should be stricter for large banks. Optimal regulation is also significantly affected by
monetary and fiscal policies. Overall, our analysis highlights the importance of considering
the interactions among different individual regulatory polices in designing
3:00-4:00pm, 796 COE
Professor Jeanne Kowalski,
3:00-4:00pm, Sep. 27, 796 COE
Professor Yijuan Hu,
Department of Biostatistics and Bioinformatics, Emory University
Meta-Analysis of Gene-Level Associations for Rare Variants Based on Single-Variant Statistics
Meta-analysis of genome-wide association studies (GWAS) has led to the discoveries of many
common variants associated with complex human diseases. There is a growing recognition
that identifying causa rare variants also requires large-scale meta-analysis. The fact that
association tests with rare variants are performed at the gene level rather than at the variant
level poses unprecedented challenges in the meta-analysis. First, different studies may adopt
different gene-level tests, so the results are not compatible. Second, gene-level tests require
multivariate statistics (i.e., components of the test statistic and their covariance matrix),
which are difficult to obtain. To overcome these challenges, we propose to perform gene-
level tests for rare variants by combining the results of single-variant analysis (i.e., p-values
of association tests and effect estimates) from participating studies. This simple strategy is
possible because of an insight that multivariate statistics can be recovered from single-variant
statistics, together with the correlation matrix of the single-variant test statistics, which can be
estimated from one of the participating studies or from a publicly available database. We show
both theoretically and numerically that the proposed meta-analysis approach provides accurate
control of the type I error and is as powerful as joint analysis of individual participant data.
This approach accommodates any disease phenotype and any study design and produces all
commonly used gene-level tests. An application to the GWAS summary results of the Genetic
Investigation of ANthropometric Traits (GIANT) consortium reveals rare and low-frequency
variants associated with human height. The relevant software is freely available.
September 13, 3:00-4:00pm, 796 COE
Professor Yao Xie,
H. Milton Stewart School of Industrial and Systems Engineering,
Georgia Institute of technology
High-Dimensional Change-Point Detection
How do we quickly detect small solar flares in a large video stream generated by NASA satellites? How do we improve detection by efficient representation of high-dimensional data that is time-varying? Besides astronomical imaging, high-dimensional change-point detection also arises in many other applications including computer network intrusion detection, sensor networks, medical imaging, and epidemiology. In these problems, each dimension of the data is obtained by a sensor, and there are multiple sensors monitoring the emergence of a signal---an abrupt change in the distribution of the observations. The goal is to detect such a signal as soon as possible after it occurs, and make as few false alarms as possible.
Two key challenges in high-dimensional change-point detection are 1) how to extract useful statistics, 2) how to find an efficient representation of the data. Many high-dimensional data exhibit low-dimensional structures such as sparsity, or the data may lie on a low-dimensional manifold. The approach I take is to exploit these low-dimensional structures in change-point detection. I will describe a mixture procedure that exploits sparsity, and MOUSSE, an online algorithm for tracking the evolving data manifold and extracts efficient statistics for change-point detection.