Statistics and Data Science Seminar: "A change point detection method in high dimensions based on U-statistics"

Speaker: B. Cooper Boniece, Utah University

Abstract: Detecting potential changes in a sequence of data is a general statistical problem that appears in a variety of scientific fields. As is common in high-dimensional statistical contexts, many classical approaches to this problem suffer from theoretical and/or practical drawbacks when the dimension of the observed data is comparable to or potentially much larger than the sample size, even under idealized independence assumptions.

In this talk, I will discuss some recent and ongoing work concerning a change-point detection method that retains favorable asymptotic properties in a high-dimensional asymptotic regime and will illustrate some of its advantages and disadvantages compared to existing approaches in the literature.  This talk is based on joint work with Lajos Horváth and Peter Jacobs.

Host: Jose Figueroa-Lopez