Grants and Contracts Details
Highly-connected physical and social domains tend to accumulate systemic risks that prevent accurate prediction from the perspective of a single domain. Previous approaches were limited in scale, resolution, or required more stringent assumptions on the underlying data generative processes. The growth of datasets in many domains and the availability of computing power now make it possible to understand and make use of the cross-domain, cross-scale linkages under more relaxed assumptions. We propose a multi-layer dynamic system to deepen the understanding of the interconnectedness of information from different domains; to provide improved prediction of risk by incorporating the interconnectedness structure without overfitting; and to identify candidate set of risk factors for intervention with the ultimate goal to mitigate the risk. PRISM will be the first to integrate dynamic high-dimensional information from different domains across multiple scales. Our approach methodically builds in stepwise dimension reduction to avoid over-fitting and applies state-of-art high-dimensional network tools to understand the interconnection among the risk indicators from different domains. The system has the flexibility to build adaptive ML/statistical models that incorporate the interconnected structure. It is expected to be powerful to offer more accurate risk prediction and inference
|Effective start/end date
|10/1/19 → 8/22/22
Explore the research topics touched on by this project. These labels are generated based on the underlying awards/grants. Together they form a unique fingerprint.