Title: Dynamical-Systems Estimation for Individualized Neuroscience
Abstract: Advances in neurotechnology now reveal neural activity with increasing resolution and coverage. However, the insight promised by this technology has not been fully realized due to the difficulty of mapping this data onto formal mathematical models by which to understand function. At issue is the difficulty of estimating high-dimensional nonlinear models when the underlying system states are not directly accessible. This problem is magnified in human neuroscience, where the difficulty of accessing the brain in-vivo precludes direct measurements. However, the strong dynamics that shape brain activity provide additional structure and richness to the problem that may be exploited. I will present paradigms to estimate high-dimensional models with special emphasis on person-specific models and noninvasive modalities: functional MRI, magnetoencephalography (MEG), and electroencephalography (EEG). The described solutions lie at the intersection of Bayesian Filtering and Deep Learning. I will also outline some burgeoning basic-science insights and translational applications that have resulted from this endeavor.
