Statistics Seminar at Georgia State University

Fall 2024-Spring 2025, Fridays 3:00-4:00pm, On-site/Virtual Seminar

Organizer: Yichuan Zhao

If you would like to give a talk in Statistics Seminar, please send an email to Yichuan Zhao at yichuan@gsu.edu

2:00-3:00pm, April 18, 2025, Distinguished Lecture: Location: 25 Park Place, Room 1441 , Professor Peter Song, Department of Biostatistics, University of Michigan,
Distinguished Lecture: Statistical Analysis of Weak Signals

Abstract: The statistical analysis of weak signals (SAWS) is a fundamental challenge in various practical domains, including questionnaire items, agrochemical residues in food, genetic variants in DNA, daily physical activity, and virus detection in wastewater. In regression analysis, identifying individual associations of weak signals is often difficult due to limited sample sizes. As a result, signals are frequently grouped into bundles to enhance detectability. Supervised homogeneity pursuit is a popular approach for forming such bundles to achieve stronger associations with outcomes of interest. Recently, we proposed a novel SAWS framework that leverages mixed-integer optimization to simultaneously perform bundle formation, association estimation, and inference. A technical innovation pertains to the reformulation of a grouping/clustering analysis as an estimation problem. This talk will discuss both the theoretical foundations and numerical performance of this approach.
3:00-4:00pm, January 31, 2025, Statistics Seminar, Location: 25 Park Place, Room 1441 , Dr. Chao Huang, Department of Epidemiology & Biostatistics, College of Public Health, University of Georgia,
Distribution-on-scalar Single-index Quantile Regression Model for Handling Tumor Heterogeneity

Abstract: This talk introduces a distribution-on-scalar single-index quantile regression modeling framework to investigate the relationship between cancer imaging responses and scalar covariates of interest while tackling tumor heterogeneity. Conventional association analysis methods assume the imaging responses are well-aligned after some preprocessing steps. However, this assumption is often violated in practice due to imaging heterogeneity. Although some distribution-based approaches are developed to deal with this heterogeneity, major challenges have been posted due to the nonlinear subspace formed by the distributional responses, the unknown nonlinear association structure, and the lack of statistical inference. Our method can successfully address all the challenges. We establish estimation and inference procedures for the unknown functions in our model. The asymptotic properties of both estimation and inference procedures are systematically investigated. The finite-sample performance of our proposed method is assessed by using both Monte Carlo simulations and a real data example on brain cancer images from TCIA-GBM collection.
2:00-3:00pm, January 31, 2025, Colloquium, Location: 25 Park Place, Room 1441 , Dr. Chi-Kuang Yeh, Department of Mathematics and Statistics, McGill University, Canada
Methods for Dependent Functional Data in Biomedical and Industrial Applications

Abstract: Functional data analysis has gained prominence with the increasing availability of complex, high-dimensional data observed continuously over time. Sequentially observed functional data (FD), referred to as functional time series (FTS), pose unique challenges in capturing and modeling serial dependencies, particularly in the presence of outliers or irregular data patterns. In this talk, I present spherical autocorrelation, a method for measuring serial dependence in FTS that examines angles between lagged functional pairs projected onto a unit sphere. By capturing both the direction and magnitude of dependence, this approach addresses limitations of traditional autocorrelation measures while maintaining robustness to atypical curves in the data. The asymptotic properties of the proposed estimators are established, enabling the construction of confidence intervals and portmanteau tests for white noise. Simulation studies validate the method’s effectiveness, and applications to model selection for daily electricity price curves and volatility measurement in densely observed asset prices demonstrate its versatility. The talk concludes with a discussion of potential extensions, including applications to multivariate FD, as well as future directions for other ongoing projects inspired by real-world challenges.
2:00-3:00pm, January 29, 2025, Colloquium, Location: 25 Park Place, Room 1441 , Dr. Hang Zhou, Department of Statistics, UC Davis,
Random Objects: Distance Profiles and Conformal Prediction

Abstract: Random objects are complex random variables taking values in general metric spaces. Although such data are increasingly common in scientific research, current statistical methodology and theory remain limited. The primary challenge in analyzing such data lies in the absence of vector space operations, such as addition, subtraction, scalar multiplication, and inner products, which are fundamental tools in conventional statistical methodologies. This talk explores object data with distance profiles and their application to conformal prediction. We introduce conditional profile average transport costs by comparing distance profiles through the optimal transport. A novel score function for random objects is proposed, enabling the construction of prediction sets using the split conformal algorithm. We develop a theoretical framework to establish uniform convergence rates for the local linear estimator involving function classes defined on metric spaces and the asymptotic conditional validity of the prediction sets. The practical utility of our proposed methodology is demonstrated through applications to network data from New York taxi trips and compositional data from brain imaging studies.
2:00-3:00pm, January 20, 2025, Colloquium, Location: 25 Park Place, Room 1441 , Dr. Shushan Wu, Department of Statistics, University of Georgia,
Subsampling in Large Graphs Using Ricci Curvature for Spatial Transcriptomics Analysis

Abstract: