CRII: CIF: Dynamic Network Event Detection with Time-Series Data

网络数据在现实世界的应用中无处不在，例如，通信网络、社交网络、计算机网络、传感器网络、电网、万维网和物联网。在这些应用中，收集的数据通常与网络结构相关联，并且网络结构可以对应于无线网络中的通信链路、社交网络中的信息流以及网络节点之间的因果关系。与个体数据相比，网络数据由于其相互关联的性质，包含了丰富的因果和相关信息，为统计推断提供了复杂性和机会。动态事件一发生就及时检测是一个非常感兴趣的问题，特别是当在真正大规模的网络中可能出现错误警报时，并且从网络的一部分进行的测量可能无法准确地捕获网络的另一部分中的事件。该项目解决了这一挑战，并开发了一个全面的框架，动态事件检测网络与时间序列数据。该项目中开发的方法可以使广泛的应用受益，例如，计算机网络中的入侵检测、流行病检测、地震事件检测以及社交网络中的假新闻检测。该项目将大大推进对如何准确建模和顺序检测具有动态性质的事件的理解，以及如何利用网络拓扑进行可靠和计算效率高的网络检测。在该项目中，将探讨以下两个重点：（一）检测单个节点的动态;（二）检测网络中的动态。实际应用，例如，将研究电动机故障检测、社交网络动态社区检测和地震事件检测，以验证本项目中开发的算法。来自概率论、信息论和随机优化的工具将被用于开发新的方法，以解决该项目中的潜在挑战。该奖项反映了NSF的法定使命，并被认为值得通过使用基金会的知识价值和更广泛的影响审查标准进行评估来支持。

Network data is ubiquitous in real-world applications, e.g., communication networks, social networks, computer networks, sensor networks, power grids, World Wide Web, and Internet of Things. In these applications, the collected data are usually associated with a network structure, and the network structure may correspond to communication links in wireless networks, information flows in social networks, and causal relationships between the network nodes. Compared to individual entities, network data, by their interconnected nature, contain rich causal and correlation information, and present both complication and opportunity for statistical inference. Timely detection of dynamic events as soon as they occur is a problem of great interest, especially as false alarms are possible in networks of really large sizes, and measurements taken from one part of the network may not accurately capture events in another part of the network. This project addresses this challenge and develops a comprehensive framework for dynamic event detection in networks with time-series data. The developed methodology in this project can benefit a wide range of applications, e.g., intrusion detection in computer networks, epidemic detection, seismic event detection, and fake news detection in social networks. This project will substantially advance the understanding of how to accurately model and sequentially detect an event with a dynamic nature, and how to exploit network topology for reliable and computationally efficient detection in networks. In the project, the following two thrusts will be explored: (i) detection of dynamics at a single node; and (ii) detection of dynamics in the network. Practical applications, e.g., fault detection in electric motor, dynamic community detection in social networks and seismic event detection, will be studied to validate algorithms developed in this project. Tools from probability theory, information theory and stochastic optimization will be used to develop novel methodologies that address the underlying challenges in this project.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.