Sophisticated functional connections appears in experiments designed for assessing the visual working memory of the human brain. Elucidating the nature of these interactions is supposedly a doorway for fruitful inference on brain cognitive processes. A challenging aspect is that they may involve brain waves oscillatory components of possibly different frequencies. We develop a statistical methodology to study collections of replicated spatially localized EEG time series. In the pursue of realistic and tractable time series models, we introduce a new class of spatio-temporal processes that are time-harmonizable. We consider available replicates and seek for the estimation of both the spatial Loeve- spectrum and the spatial dual-frequency coherence function. In other words we measure a squared correlation coefficient across different frequencies between any spatial locations. Our method exploit the spatial correlations to improve the local estimation of the coherence between different frequency bands. We derive the asymptotic distribution of the spatial coherence function and build bootstrap-based confidence intervals. This is based on joint work with John Aston, Dominique Dehay, Anna Dudek and Denes Scusz.
Dernière modification le 7 Mars 2017