Stat Colloquium: Dr. Taposh Banerjee

Title: Quickest Change Detection Algorithms for Experiment Design, Nonstationary Data, and High-Dimensional Data

Abstract: Anomaly detection is a common problem in science and engineering. In the statistical subject of sequential analysis, this problem is formulated as the problem of quickest change detection (QCD). In the QCD problem, a decision-maker collects a sequence of observations. At some point in time, called the change point, the statistical behavior of the data changes. The goal of the QCD problem is to detect this change with the minimum possible delay, subject to a constraint on the rate of false alarms. The quickest change detection problem has applications in sensor networks, statistical quality control, neuroscience, medicine, stochastic control, and cyber-physical systems. I will first introduce the audience to the fascinating history of this problem and discuss important results in this area. I will then briefly discuss three of my recent contributions in this area: QCD with cost-constraint experiment design, QCD for multistream nonstationary data, and a score-based approach to QCD in high-dimensional data.