Equation-free, data-driven modeling for large-scale neural recordings


November 16, 2016 - 1:00pm - 2:00pm
Northwest 243
About the Speaker
Bingni Brunton
Speaker Affiliation: 
University of Washington


Our ability to acquire high-dimensional recordings of the brain, sampling both larger and more detailed neuronal networks, presents a growing challenge for computational modeling and analysis. In recent years, data-driven, equation-free techniques are gaining substantial traction in application to complex systems; these techniques construct models on observed measurement data directly, making them well suited to describe neural systems, for which the governing equations are partially known at best. I will talk about recent work adapting one data-driven dynamic modeling algorithm known as dynamic mode decomposition (DMD), originally developed for studying fluid physics, to large-scale neural recordings. DMD describes high-dimensional time-series data using coupled spatio-temporal modes to build a dynamic model. The interface between dynamical systems theory and modern big data enables some unique opportunities, including connections to nonlinear dynamics and control theory.