Collaborative Research: ATD: a-DMIT: a novel Distributed, MultI-channel, Topology-aware online monitoring framework of massive spatiotemporal data

Duke University, Durham NC

Investigators

Abstract

Given current technological, societal and environmental changes, we face ever more diverse potentially destructive threats. Technology has enabled the collection of massive spatiotemporal datasets, which can be used for real-time identification of potential threats. The complexities of such data create exciting challenges for online threat detection, involving the learning and integration of complex nonlinear embeddings for efficient monitoring, the fusion of multiple spatiotemporal data sources for improving detection performance, and scalability for real-time implementation on distributed computing systems. This project will develop a novel Distributed, MultI-source, Topology-aware (a-DMIT) online threat detection framework that tackles these challenges for massive, high-dimensional spatiotemporal data. In developing reliable, scalable and versatile threat detection methods (supported by theory and algorithms), a-DMIT has the potential to improve national health and defense in a broad range of areas, including environmental monitoring, crime monitoring and mobile health. The a-DMIT project will contribute to education by involving undergraduate and graduate students in the research, and developed software will be made publicly available. a-DMIT will develop three new detection methods that jointly tackle fundamental challenges in online monitoring of massive data streams. The first method, called PERsistence diagram-based ChangE-PoinT detection (PERCEPT), is a novel non-parametric, topology-aware algorithm that extends state-of-the-art tools in topological data analysis for efficient monitoring of high-dimensional data streams. The second, called MUlti-source Monitoring via Gaussian Processes (MUM-GP), is an efficient online Bayesian non-parametric detection method for multi-source spatiotemporal data. The third, called Conditional Auto-Regressive Distributed (CARD) detection, is an online spatiotemporal network monitoring procedure that leverages neighboring spatial information in a distributed and decentralized fashion. a-DMIT will be usable for a wide range of modern threat detection applications, including environmental monitoring, crime monitoring, satellite image monitoring and power grid security. This award reflects NSF's statutory mission and has been deemed worthy of support through evaluation using the Foundation's intellectual merit and broader impacts review criteria.