ATD: Collaborative Research: Adaptive and Rapid Spatial-Temporal Threat Detection over Networks

该项目旨在开发创新的机器学习和统计算法，用于检测、预防和响应网络威胁。两个具体的应用是监测来自美国各地诊所网络的多重耐药(MDR)淋病的威胁，以及监测艾滋病病毒在患者群中的传播。这项研究在许多其他实际应用中都有影响，包括生物监控、工程、国土安全、金融和公共卫生，在这些应用中，收集大规模时空数据流的目的是快速检测和预防威胁。这项研究旨在开发关键的可扩展算法和方法，以有效和高效地监控、分析和优化在这些情况下的响应。此外，该项目将通过将研究成果注入课程并让博士生参与研究来整合研究和教育。该项目旨在开发创新的算法，通过将时空模型、常微分方程(ODE)模型与变点检测相结合，以及在网络上监控大规模时空数据时的多臂强盗和集成方法来实现快速威胁检测。具体而言，在三个相互关联的研究任务中开发了高效的可扩展算法，包括：(1)通过将“背景+异常+噪声”分解框架与顺序变化点检测相结合来快速检测威胁；(2)通过在变化的环境中应用多臂强盗算法和自适应采样来对威胁进行预测性分析，以评估人口水平上不断增加的风险；以及(3)通过基于校准的和数据驱动的时空模型开发嵌套的集成模型来对威胁进行规范分析，以便更好地评估预防/干预措施的效果。该项目的成果预计将显著推进时空模型、在线学习、流数据分析和大规模推理方面的最先进水平。该奖项反映了NSF的法定使命，并通过使用基金会的智力优势和更广泛的影响审查标准进行评估，被认为值得支持。

This project aims to develop innovative machine learning and statistical algorithms for detecting, preventing, and responding to threats over networks. Two concrete applications are monitoring the threat of multi-antibiotic-resistant (MDR) gonorrhea from a network of clinics across the United States and monitoring HIV transmission in clusters of patients. The research has impact in many other practical applications, including biosurveillance, engineering, homeland security, finance, and public health, where large-scale spatial-temporal data streams are collected with the aim of rapid detection and prevention of threats. The research aims to develop crucial scalable algorithms and methods to effectively and efficiently monitor, analyze, and optimize responses in these situations. In addition, the project will integrate research and education by infusing the research findings into the curriculum and by involving Ph.D. students in research. This project aims to develop innovative algorithms for rapid threat detection by combining spatial-temporal models, ordinary differential equation (ODE) models with change-point detection, and multi-armed bandit and ensemble methods when monitoring large-scale spatial-temporal data over networks. In particular, efficient scalable algorithms are developed in three interrelated research tasks, including (1) rapid detection of threats by combining a "background + anomaly + noise" decomposition framework with sequential change-point detection; (2) predictive analytics of threats by applying multi-armed bandit algorithms and adaptive sampling in the changing environments to assess increasing risks at the population level; and (3) prescriptive analytics of threats by developing nested ensemble models based on calibrated ODE and data-driven spatial-temporal models so as to better assess the effects of prevention/intervention actions. Results of the project are expected to significantly advance the state of the art in spatial-temporal models, online learning, streaming data analysis, and large-scale inference.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.