Title: Heavy-Tailed Mixture Models in Neuroscience
Abstract:
In this talk, I will present new methods for analyzing heavy-tailed data in neuroscience applications, focusing on brain rhythm signals, which are key markers of brain function linked to various behavioral and cognitive states. Measured by EEG, these signals span a range of frequency bands, such as alpha (8–13 Hz) and beta (13–20 Hz), which have shown heavy-tailed properties. Recognizing that traditional random probability measures like the Dirichlet process (DP) struggle with heavy tails, our first contribution is to characterize the tail behavior of the normalized generalized gamma (NGG) process, showing that it maintains heavy tails when the centering distribution is heavy-tailed — unlike the DP. Our second contribution introduces two classes of heavy-tailed mixture models, including multivariate and predictor-dependent extensions, to assess covariate effects on multivariate heavy-tailed responses. Simulations highlight the models effectiveness, and we apply them to EEG data from participants who experienced varied stimuli.
There will be brief introductions and AGM formalities led by local group chair Dr Georgios Karagiannis, followed by a talk by
Dr Vianey Palacios Ramirez, Newcastle University.