We develop time-frequency methods and models for characterizing, estimating and comparing dependence in multivariate non-stationary time series. This is motivated by a visual-motor experiment where the goal is to study differences in brain connectivity between EEG channels across different experimental conditions. We shall characterize dependence using a variety of time-frequency measures such as evolutionary partial coherence, partial cross-correlation and mutual information. We shall develop these measures under non-stationary time series models that are based on localized transforms. We shall develop shrinkage-based methods for estimation and randomization tests for comparing brain network across different experimental conditions.
This work has been in collaboration with Mark Fiecas at Brown University.