CAREER: Streaming Data Analysis in Sensor Networks

本研究旨在为传感器网络中的流数据分析提供统计基础和一系列有效的可扩展方法。在许多应用中，传感器网络被部署到随时间和空间变化的环境的在线监测，其目标是早期检测一些可能造成重大损害的特定触发事件。然而，来自分布式的、不同来源的流数据的性质和受限的网络资源（在通信、计算、成本、原始数据的隐私等方面），这些挑战要求开发新的统计工具、方法和理论。在这个项目中，研究人员提出了一种新的一般框架，用于监测传感器网络，其中触发事件可能会影响不同的传感器或数据流不同。一些具体的研究主题包括纯（共识或并行）检测和检测后的推理，在不同的场景下，这取决于传感器观测模型和传感器协议的设计要求。此外，还将通过将研究成果融入课程、举办研讨会和讲习班、为研究生和本科生提供咨询等方式，将研究与教育结合起来。Senor网络在健康和环境监测、生物医学信号处理、无线通信、计算机网络入侵检测、生物监视等领域有着广泛的实际应用。一方面，该研究项目将提供重要的统计工具，以有效地监测和分析这些传感器网络应用中的动态数据流。另一方面，它在这些应用中也有令人沮丧但深刻的影响：面对所提出的研究的（渐近）最优性理论所隐含的限制，从业者和研究人员可能需要不断寻找更好的数据源，以在其特定应用中实现所需的系统性能，而不是依赖于现有数据源的改进方法。

This research aims to offer statistical foundation and a host of efficient scalable methodologies for streaming data analysis in sensor networks. In many applications, sensor networks are deployed to online monitoring of changing environments over time and space, with a goal of early detection of some particular trigger events that can cause significant damage. However, the nature of streaming data from distributed, diverse sources and the constrained network resources (on communication, computing, costs, privacy of raw data, etc.) pose significant challenges, which require the development of new statistical tools, methods and theories. In this project, the investigator proposes a novel general framework for monitoring sensor networks in which a trigger event may affect different sensors or data streams differently. Some specific research topics include pure (consensus or parallel) detection and inference after detection, under different scenarios, depending on the models for sensor observations and the design requirements of sensor protocols. In addition, the research will integrate research and education by infusing the research findings into the curriculum, by organizing seminars and workshops, and by advising graduate and undergraduate students.Senor networks have broad real-world applications, including but not limited to health and environmental monitoring, biomedical signal processing, wireless communication, intrusion detection in computer networks, and biosurveillance. On the one hand, this research project will offer crucial statistical tools to effectively and efficiently monitor and analyze dynamic data streams in these sensor network applications. On the other hand, it also has a frustrating yet profound implication in these applications: Faced with the limitations implied by the (asymptotic) optimality theories of the proposed research, practitioners and researchers may need to constantly look for better data sources to achieve desired system performance in their specific applications rather than relying on an improved methodology for existing data sources.