Many important big-data problems can be modeled as networks in which nodes are mapped in physical space. To address these problems, the project investigates a novel network abstraction called a spatial-relational network. This abstraction targets a broad class of applications with unique big-data challenges, including financial fraud, epidemiology, and systems engineering. The distinguishing feature of spatial-relational networks is that the network component models relationships that correspond to fundamentally non-spatial modes of influence and information flow, whereas the spatial relationships represent a different type of heterogeneous interaction. As an example, in the human connectome, edges correspond to regions of the brain that fire together, whereas spatial proximity reflects patho-physiological aspects such as the existence of a lesion. This project aims to develop models, methods, and software for important analysis problems on spatial-relational networks as well as validate them on a range of real-world problems. The project enables fundamentally new analytic techniques in many applications while significantly enhancing the accuracy and efficiency of analyses techniques in others. These range from novel fraud detection algorithms in online transactions to study of onset and progression of neurodegenerative diseases such as Alzheimer's and Parkinson's. Beyond its research and applications' impact, the project also targets a number of educational contributions. These include development of a curriculum for data sciences, online modules on big-data analytics, workshops and tutorials on the spatial-relational network abstraction at major conferences, and opportunities for undergraduate research.
This project establishes the foundations of spatial-relational networks as a fundamental and scalable big-data abstraction that enables novel analyses. It aims to derive efficient and effective computational representations for spatial-networks, high-performance algorithms for spatial-relational networks, modeling and analysis of dynamic spatial-relational networks in which the node relations and attributes are dynamic, but their spatial locations are fixed. Project goals include software development and comprehensive validation on selected applications in the analysis of data from the Human Connectome Project, the Allen Brain Atlas, and location-based social networks.
