Activity Number:
|
157
- Contributed Poster Presentations: Section on Statistics in Imaging
|
Type:
|
Contributed
|
Date/Time:
|
Monday, August 8, 2022 : 10:30 AM to 12:20 PM
|
Sponsor:
|
Section on Statistics in Imaging
|
Abstract #323324
|
|
Title:
|
Hierarchical Bayesian Nonparametric Decomposition of Multiple Multivariate Brain Signals
|
Author(s):
|
Guillermo Cuauhtemoctzin Granados Garcia* and Raquel Prado and Hernando Ombao
|
Companies:
|
King Abdullah Universe of Science and Technology and University of California Santa Cruz and King Abdullah University of Science and Technology
|
Keywords:
|
MCMC;
Bayesian nonparametrics;
EEG;
Nested Dirichlet Process;
Hierarchical model
|
Abstract:
|
Understanding the dynamic dependence between multivariate brain signals is crucial for modern neuroscience research. A nonparametric Bayesian model is proposed to decompose observed signals into quasi-periodic processes via second-order autoregressive processes representing shared waveforms across the signals. The multivariate signals dependence structure is inferred via a Nested Dirichlet process model that models the data as a mixture of an unknown number of latent processes and clusters the primary waveforms across nodes. The hierarchical model is extended to infer group-level features used to compare two populations. A Metropolis within Gibbs algorithm is implemented to compute the posterior distributions of the parameters mixed with optimization strategies to improve computational performance. The model effectiveness is demonstrated in a simulation study of two groups with smooth and abrupt changes over the behavior and dependency structure of the simulated signals waveforms. The Nested DP approach is used to compare the dependence structure and shape of their brain signals waveforms of a group of alcoholics and control subjects during a visual recognition experiment.
|
Authors who are presenting talks have a * after their name.