Characterising spatio-spectral subject differences in fMRI and EEG using ML techniques

CECILIA JARNE; DIEGO VIDAURRE

Tokio

Otro; RIKEN CBS Summer Program; 2023

Riken Institute

Electrophysiological brain signals exhibit oscillatory and aperiodic activity linked to physiological,cognitive, behavioural, and disease states. Such activity is crucial in coordinating interregional information transference. Predicting optimally different features from brain data remains an open question. A method to predict features based on frequency content across regions is presented.Kernel embeddings of distributions (KME) map probability distributions to vectors in high-dimensional feature space, completely characterizing the distribution. Maximum Mean Discrepancy (MMD) characterizes differences in such distributions, quantifies similarities between subjects, and we can relate them to brain age.