Distributed model building and recursive integration for big spatial data modeling

Area of Science:

• Computational neuroscience
• Statistical modeling
• Neuroimaging analysis

Background:

• Neuroimaging studies require computationally efficient spatial methods.
• Analyzing ultra-high-dimensional data in Gaussian process models is challenging.
• Existing methods may lack tractability for large-scale neuroimaging datasets.

Purpose of the Study:

• To develop a distributed and integrated framework for Gaussian process model parameter estimation and inference.
• To address computational challenges in neuroimaging studies with ultra-high-dimensional likelihoods.
• To enable new insights into autism spectrum disorder using advanced spatial analysis.

Main Methods:

• A distributed model-building approach focusing on local data perspectives.
• An integrated estimation and inference procedure for recursively partitioned spatial domains.
• Theoretical investigation and simulation studies to validate statistical and computational properties.

Main Results:

• The proposed framework offers computational and statistical efficiency.
• The integration procedure effectively handles dependence within and between spatial resolutions.
• The approach is validated through theoretical analysis and simulations.

Conclusions:

• The developed framework provides a computationally tractable solution for spatial neuroimaging analysis.
• This approach facilitates robust estimation and inference in complex Gaussian process models.
• The framework enables novel discoveries in autism spectrum disorder research using the Autism Brain Imaging Data Exchange.