Statistics Seminar Series: Pramita Bagchi (GW School of Public Health)

About

Please join the GW Statistics Department for a talk featuring Pramita Bagchi, Department of Biostatistics and Bioinformatics at the Milken Institute School of Public Health, George Washington University.

Adaptive Frequency Band Learning of Nonstationary Functional Time Series: An Application to High-Dimensional EEG Signals

Abstract: Frequency-domain properties of high-dimensional electroencephalogram (HD-EEG) signals contain valuable information about brain activity and functioning. These properties tend to vary over time and are characterized through the time-varying spectral density. Rather than analyzing the time-varying spectral density directly, biomedical researchers work with low-dimensional summary measures by averaging power within pre-specified frequency bands. Partitioning frequencies into bands significantly affects how well frequency band summary measures characterize the underlying spectral density, making it an important yet often overlooked analytical consideration. HD-EEGs represent an important class of biomedical signals that can be appropriately characterized as nonstationary functional time series. We develop a method for learning frequency band structures for functional time series that optimally summarize time-varying dynamics. We develop a scan statistic and scalable search algorithm to detect changes in the frequency domain for adaptive frequency band estimation. We validate its use in practice by establishing essential theoretical properties and demonstrating accurate frequency band estimation through numerous simulation studies. In a study of participants alternating between eyes-open and eyes-closed conditions, the method accurately recovers complex time-varying dynamics of HD-EEGs consisting of spatially localized nonstationary behavior and synchronization in power across frequencies from different pre-specified bands used in practice.

About the Speaker: Pramita Bagchi is an associate professor in the Department of Biostatistics and Bioinformatics at the George Washington University Milken Institute School of Public Health. She received her PhD in statistics from the University of Michigan and completed postdoctoral training at Ruhr University Bochum in Germany. Her research focuses on statistical inference for data with complex dependence, with interests spanning functional data analysis, time series, spatial statistics and topological data analysis. Her collaborative research includes applications in climate modeling, environmental health, genomics and clinical research.