Title: Sparse plus Low-rank Graphical Models of Time Series for Functional Connectivity in MEG
Advisors: Emily Fox and Rajesh Rao
Abstract: A fundamental problem in neuroscience is inferring the functional connections between brain regions that underlie cognitive behaviors such as vision, speech, and audition. Magnetoencephalography (MEG) has become a popular neuroimaging technique for studying these functional connectivity networks. However, unlike typical approaches for learning these networks from data, we treat the MEG signals as time series rather than independent observations. We represent the functional connectivity network through conditional independence statements between signals on the cortical surface, encoded via a graphical model of time series. Importantly, we extend previous techniques for learning graphs of time series by incorporating a low-rank component, accounting for latent signals that would otherwise lead to inferring spurious connections. We evaluate the model on synthetic data as well as real MEG data collected from an auditory attention task.