Sparsity in the multivariate wavelet framework: A comparative study using epileptic electroencephalography data

摘要

We consider how recently developed multi-resolution exploratory graphical models (MR-EGM) may be estimated in a practical real-world situation. A simple cross-validation procedure based on minimising predictive risk is presented as a means to estimate tuning parameters. Through the use of electroencephalography (EEG) data, we attempt to use such a procedure to build a generative (multi-resolution) model of the electrical dynamics in the brain throughout an epileptic seizure. Brain dynamics are analysed by projecting the estimated model parameters onto their principle components where we identify two clusters of seizure activity. To conclude, we discuss the interpretation of such a principle component analysis and how well we can generalise between seizures on a specific patient.