Bayesian Nonparametric Multivariate Mixture of Autoregressive Processes: With Application to Brain Signals

Abstract: One of the goals of neuroscience is to study interactions between different brain regions during rest and while performing specific cognitive tasks. The Multivariate Bayesian Autoregressive Decomposition (MBMARD) is proposed as an intuitive and novel Bayesian non-parametric model to represent high-dimensional signals as a low-dimensional mixture of univariate uncorrelated latent oscillations. Each latent oscillation captures a specific underlying oscillatory activity and hence will be modeled as a unique second-order autoregressive process due to a compelling property that its spectral density has a shape characterized by a unique frequency peak and bandwidth, which are parameterized by a location and a scale parameter. The posterior distributions of the parameters of the latent oscillations are computed via a metropolis-within-Gibbs algorithm. One of the advantages of MBMARD is its robustness against misspecification of standard models which is demonstrated in simulation studies. The main scientific questions addressed by MBMARD are the effects of long-term abuse of alcohol consumption on memory by analyzing EEG records of alcoholic and non-alcoholic subjects performing a visual recognition experiment. The MBMARD model exhibited novel interesting findings including identifying subject-specific clusters of low and high-frequency oscillations among different brain regions.