ATD: Multiscale Anomaly Detection in Spatio-Temporal Multilayer Networks Encoding Human Mobility

来自匿名移动电话设备的人类流动性数据变得越来越可用，从而能够检测人类活动的异常和潜在威胁。由于数据量大，很难检测到在不同空间和时间分辨率上出现的异常。此外，人员流动往往与不同类别的地方(如杂货店和学校)联系在一起，在一个类别中明显的异常现象在另一个类别中可能不明显。该项目将为将不同类别的人类运动编码为时空多层网络提供新的数学模型和理论。它将开发新的算法来检测运动模式异常，这可以帮助更好地预警涉及国家安全的异常事件。它还将增进我们对异常和灾难性事件对不同类别的人类行动的影响的理解。这个项目将通过支持研究生来促进教育，并将促进数学和地理之间的跨学科研究。将实施和公开共享开源软件工具，以帮助研究人员和决策者更好地预测、检测和计划对未来异常事件的反应。为了表示不同类别的人体运动并有效地检测相关异常，调查人员将在这个项目中追查三个目标。Target 1将开发对美国匿名手机位置数据进行编码的时空多层网络模型。将广泛研究与正常和预期异常情况(例如，假期)相关的网络结构属性，以开发一系列现实的生成模型。目标2将基于多层网络模型，使用无监督频谱算法，建立检测和表征运动模式异常的算法。随机矩阵理论将被用来获得关于如何最佳地对数据进行预处理以最大限度地提高不同空间和时间分辨率的异常的“可探测性”的理论指导方针。目标3将把开发的多层网络模型和异常检测算法应用于两个案例研究，以进一步完善我们的模型和算法，并对这两个重要事件的演变和影响获得新的见解。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。

Human mobility data from anonymized mobile phone devices are becoming increasingly available, enabling the detection of anomalies and potential threats from human movements. Due to the massive scale of data, it is difficult to detect anomalies that occur at different spatial and temporal resolutions. In addition, human movements are often associated with different categories of places (e.g., grocery stores and schools), and anomalies that are evident in one category may be obscure in another. This project will furnish new mathematical models and theories for encoding different categories of human movements as spatial-temporal multilayer networks. It will develop new algorithms to detect movement-pattern anomalies, which can help better forewarn anomalous events concerning national security. It will also advance our understanding of the impacts of anomalous and disastrous events on different categories of human movements. This project will contribute toward education by supporting graduate students and will facilitate interdisciplinary research between mathematics and geography. Open-source software tools will be implemented and publicly shared to help researchers and decision makers better predict, detect, and plan responses to future anomalous events. To represent human movements in different categories and effectively detect the associated anomalies, the investigators will pursue three targets in this project. Target 1 will develop spatial-temporal multilayer network models encoding anonymized mobile phone location data of the United States. The network-structural properties associated with both normal and expected anomalous situations (e.g., holidays) will be extensively examined to develop a family of realistic generative models. Target 2 will build algorithms to detect and characterize movement-pattern anomalies based on the multilayer network models using unsupervised spectral algorithms. Random matrix theory will be employed to obtain theoretic guidelines for how to optimally preprocess data to maximize the “detectability” of anomalies at different spatial and temporal resolutions. Target 3 will apply the developed multilayer network models and anomaly detection algorithms to two case studies, to further refine our models and algorithms and to gain new insights into the evolutions and impacts of the two important events.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.