ATD: Nonparametric Testing and Fast Computing Methods for Spatiotemporal Models with Applications to Threat Detection

点击翻译按钮获取中文摘要

The study of human dynamics aims to understand human behaviors using analytical models. It has received substantial attention in the security and defense area not only for its potential in detecting human anomalies but also for its capability in containing potential disastrous damages and mental horror in human society. The recent development in social media has revolutionized daily life and inaugurated a new era in human dynamics study. It has been demonstrated that certain human behavior can be modeled quantitatively using proxy tools. Social media data and wearable device data are two major sources of proxies that can be used to understand spatiotemporal trends from which we can identify abnormal patterns in human dynamics. The abnormal patterns can be used as an indicator for disasters. Although social media data and wearable device data contain a wealth of information to understanding human behavior, they are constantly evolving as users generate new content or as new routes are introduced. Classic statistical models are not enough to model constantly-evolving spatial and temporal trends. More importantly, the computational cost for a spatiotemporal model is extremely high, which poses a significant challenge for real-time analysis of human dynamics data. To overcome these challenges, we propose novel statistical theory, methods, algorithms for efficiently analyzing local and global trends of a spatiotemporal model. The project will train students to participate in cutting-edge and interdisciplinary big data research. Despite the urgent need, statistical tools for human dynamic studies are still lacking. In this project, we aim to develop spatiotemporal models to understand the inherent spatial/temporal trends of social media and personal wearables data. The key challenge for analyzing large-scale spatiotemporal data is the super-large sample size. For example, every day Twitter takes in hundreds of million tweets. Many of these tweets are recorded in public streams. We develop scalable computational methods to surmount the challenge. The fast, in-network computing principles and middleware developed in this project are fundamental and indispensable tools for "big data" computation and autonomous systems. The proposed framework can be used to discover unusual events in any super-large dynamic data set and inspire a new line of research in big data analytics. The proposed statistical tools are widely applicable in science, engineering, and humanities. The research will be conducted in collaboration with experts in both geography and statistics; consequently, the proposed work will be informed by rapid empirical feedback and can incorporate modern advances in spatiotemporal modeling.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.