All University of South Carolina system institutions will be closed through the end of the spring semester. Columbia campus virtual instruction will continue through the end of the Summer II semester (Aug 1). Details can be found on the coronavirus landing page.

# Department of Statistics

## Karl Gregory

 Title: Assistant Professor Department: StatisticsCollege of Arts and Sciences Email: gregorkb@mailbox.sc.edu Phone: 803-777-3859 Office: LeConte 219C Resources: My WebsiteCurriculum Vitae [pdf]Department of Statistics

#### Research

My dissertation work centered on bootstrap methodology for time series data and on a two-sample testing problem in which the number of variables exceeded the sample size: The goal of bootstrapping is to estimate the sampling distribution of a sample quantity; the idea is to simulate repeated sampling from the population by repeatedly drawing samples from the sample itself. In the time-series setting, one must be careful to make sure that the resampling scheme preserves the dependence structure of the data. The two-sample problem arose in collaboration with colleagues at the University of Texas MD Anderson Cancer Center and concerned the detection of differences in copy number variation proﬁles (deletions or duplications of sequences of DNA along the chromosome) between long- and short-surviving glioblastoma multiforme patients.

During a two-year post-doc at the Universities of Mannheim and Heidelberg, I began studying nonparametric regression in the high-dimensional setting—when there are a large number of predictors relative to the sample size and each predictor inﬂuences the response through a function of unspeciﬁed form. In the meantime, I also worked on bootstrap methods for quantile regression in time series. My dissertation work centered on bootstrap methodology for time series data and on a two-sample testing problem in which the number of variables exceeded the sample size: The goal of bootstrapping is to estimate the sampling distribution of a sample quantity; the idea is to simulate repeated sampling from the population by repeatedly drawing samples from the sample itself. In the time-series setting, one must be careful to make sure that the resampling scheme preserves the dependence structure of the data. The two-sample problem arose in collaboration with colleagues at the University of Texas MD Anderson Cancer Center and concerned the detection of differences in copy number variation proﬁles (deletions or duplications of sequences of DNA along the chromosome) between long- and short-surviving glioblastoma multiforme patients.

Since coming to USC in August of 2016, I have begun to work on bootstrap methods for high-dimensional regression problems as well as penalized regression methodologies in pooled testing problems, in which pools of individuals are simultaneously tested for the presence of a disease.