Change Point Detection for High-dimensional Time Series Models With Local Dynamics

Presented by Abolfazl Safikhani, Department of Statistics, George Mason University

Friday, April 5 2024
10:00 AM-11:00 AM ET
AUST 202
Webex Meeting Link
Coffee will be served at 9:30 am in the Noether Lounge (AUST 326)

Sequential monitoring of multivariate time series to find sudden changes in the data generating process is a canonical problem in statistics and signal processing. Most developed detection algorithms work under two main assumptions: (a) there are no cross-correlations between time series components, and (b) observations between two change points follow a stationary model with fixed parameters. Both of these assumptions are unrealistic in real data applications. Failing to include cross-correlations and dynamic/non-stationary structures may lead to over-fitting and/or inaccurate change point identification in such algorithms. In this talk, first a general modeling framework is introduced to include local dynamic structures and cross-correlations in multivariate time series. Then, a novel sequential detection algorithm is proposed to estimate the location of change points while also estimating all model parameters, including cross-covariance parameters. Theoretical properties are established under mild conditions including controlling the false positive rate, detection power calculation, and localization error bounds. Finally, empirical performance of the proposed method is investigated through various simulation settings, comparison with several competing methods, and real data examples including paper production data. This is a joint work with Yuhan Tian and Kamran Paynabar.

Speaker Bio:

Dr. Safikhani is currently an assistant professor in the department of statistics at George Mason University. He received his PhD from the department of statistics and probability, Michigan State University. Prior to GMU, he has held positions at Columbia University and University of Florida. His main research interests include network modeling, high-dimensional statistics, spatio-temporal models, change point detection, and applications in urban planning, neuroscience, and smart cities. He has published in top-tier statistical journals and machine learning conferences including JASA, Technometrics, JCGS, EJS, EJP, IEEE-TSP, and NeurIPS and his research has been supported by NSF. He is currently an Associate Editor of Statistica Sinica and Data Science in Science.