Collaborative Research: Bayesian State-Space Models for Behavioral Time Series Data

点击翻译按钮获取中文摘要

This research project will develop novel statistical models and inferential methods for the analysis of multi-domain behavioral data and time series with complex temporal and dependence structures. This research has the potential to advance the knowledge on the neural underpinnings of human and animal behavior. Neuroscience studies often involve the analysis and integration of data from different domains, such as behavioral and neural-derived data. The focus of this project will be on developing statistical methods for studying temporal data derived from functional magnetic resonance imagining (fMRI) and local field potentials, such as neural-derived brain signals. These methods also are applicable to other types of brain signals, such as electroencephalograms and magnetoencephalograms. These statistical approaches will integrate data from different domains and could be used by behavioral scientists to directly test for associations between decision making and brain response. The statistical tools that will be developed in this project are general and could be used to advance knowledge in other fields that collect temporal data with complex structure, such as sociology (network modeling), environmental sciences, linguistics, and signal processing.The project will develop Bayesian state-space models for activation and connectivity in fMRI data. These models will be used to simultaneously estimate the hemodynamic behavior in local areas of the brain and to estimate inter-dependence between brain regions in a network, while taking into account variations across subjects and differences across experimental conditions. The Bayesian state-space models and related inferential tools then will be extended to consider associations between the neural-derived brain signals and behavioral data under the context of behavioral experiments. Bayesian state-space models for brain connectivity using electrophysiological signals also will be developed. To deal with high computational demands for inference resulting from increased model complexity and massive data, the methods will be implemented using parallel computing.