Eager: Cyberattacks on Commercial IoT Networks Estimating Large Dimension Parameters for Big Data

摘要ECCS-1744129标题：EIGER：商业物联网网络的网络攻击评估对象位置非技术描述：虽然互联网已经存在多年，但将传感和控制技术集成到互联网中以产生所谓的物联网(IoT)仍然非常不成熟，并带来了危险的新的未解决的安全问题。例如，针对汽车处理器的网络攻击已经被观察到。汽车制造商正在开发关键系统，旨在融合来自几个复杂传感器的数据，以确保自动驾驶汽车避免与人和动物相撞。该项目将寻求建立一套完整的商用物联网/传感器对象位置估计、网络攻击缓解和影响的理论，其基础是：(1)严格证明在一些合理的假设下可以识别被攻击的传感器；(2)基于严格的估计理论分析被攻击系统可能达到的性能范围。这一新理论应该会带来针对智能家居、智能建筑、智能工厂和日常生活中所依赖的其他商业物联网/传感器系统的网络攻击的技术。申请的大部分资金将用于资助研究生。在这些重要的跨学科领域，将继续从代表性不足的群体中培养研究生和本科生。这项研究项目、CLASS和利哈伊的电力综合网络(INE)跨学科研究倡议之间的协调与该项目相关的更广泛的影响。研究成果将被纳入目前和未来的利哈伊班级。课堂笔记可能会演变成一门关于传感系统安全的简短课程，并可能成为一本关于传感系统安全的书，以提供巨大的教育影响。技术描述：对于正在考虑的对象定位问题，研究将准确地描述在没有攻击存在的情况下，每个传感器上的观测如何约束可能的对象位置。这项研究还将表征增加每个传感器的观测数量和没有攻击时增加传感器数量的影响。此外，研究将分析相交约束在多个传感器上的影响，以准确地展示相交约束如何在没有攻击的情况下提供包含对象位置的严格较小的集合。大偏差分析表明，当每个传感器有足够多的观测值时，该方法能够以较高的概率正确定位目标。观察次数较少的情况将使用适当的界限进行分析，这些界限对于有限数目的观察是准确的。这些分析方法将在攻击下使用，以表明显著的攻击将驱动对空集的交叉约束。大偏差分析将被用来确定该方法可以正确地识别被攻击的传感器，在一些合理的假设下，当每个传感器上有足够数量的观测数据且概率很高时。观察次数较少的情况将使用适当的界限进行分析，这些界限对于有限数目的观察是准确的。

Abstract ECCS -1744129Title: Eager: Cyberattacks on Commercial IoT Networks Estimating Object Position Non-technical description: While the internet has been available for many years, the integration of sensing and control technology into the internet to yield what is being called the Internet of Things (IoT) is still very immature and brings dangerous new unaddressed security problems. For example, cyberattacks on automotive processors have already been observed. Car manufacturers are developing critical systems aimed at fusing data from several complex sensors to ensure self-driving automobiles avoid collisions with people and animals. The proposed project will seek to develop a complete theory of commercial IoT/sensor object location estimation network attack mitigation and impact based on: (1) rigorous proofs that the attacked sensors can be identified under some reasonable assumptions and (2) rigorous estimation theory-based analysis of the possible range of performance that the attacked system can achieve. The new theory should lead to technology to protect against cyber attacks on smart homes, smart buildings, smart factories and other commercial IoT/sensor systems relied upon in daily lives. The major portion of the requested funds will go towards supporting graduate students. Educating graduate and undergraduate students, from under represented groups, in these important cross-disciplinary areas will be pursued. Coordination between this research project, classes and Lehigh's Integrated Networks for Electricity (INE) interdisciplinary research initiative is broader impact associated with this project. Research results will be incorporated into current and future Lehigh classes. Class notes might evolve into a short course, and possibly a book on security of sensing systems, to provide a large educational impact.Technical description: For the object localization problems under consideration, the research will characterize precisely on how observations at each sensor constrain the possible object position when no attacks are present. The research will also characterize the impact of increasing the number of observations per sensor and the number of sensors when no attacks are present. Further, the research will analyze the impact of intersecting constraints at multiple sensors to show exactly how the intersected constraints provide a strictly smaller set containing the object location under the case of no attacks. Large deviation analysis will be used to demonstrate that this approach can properly localize the target with high probability when a sufficient number of observations are available at each sensor. Cases with fewer observations will be analyzed using appropriate bounds that are accurate with a finite number of observations. These analytical approaches will be employed under attacks to show that significant attacks will drive the intersected constraints to the empty set. Large deviation analysis will be used to establish that this approach can properly identify attacked sensors, under some reasonable assumptions, when a sufficient number of observations are available at each sensor with high probability. Cases with fewer observations will be analyzed using appropriate bounds that are accurate with a finite number of observations.