TWC SBE: Small: From Threat to Boon: Understanding and Controlling Strategic Information Transmission in Cyber-Socio-Physical Systems

随着网络--社会--物理和基础设施系统越来越依赖数据，并整合越来越多的不同的、有时是非常规的、可能是不可信的数据源，越来越有必要考虑在存在战略和/或自利传感器的情况下的估计问题。这类问题被称为战略信息传输(SIT)，不同于经典的容错估计，因为传感器不仅出现故障或故障，而且为了自身的利益积极地试图误导估计器。这种战略行为至少可以通过两种机制发生，传感器劫持和故意误报，这两种机制都在实践中得到了观察。战略传感器和故障传感器之间的一个特别明显和根本的区别是，战略传感器的数量不一定会导致错误的估计。事实上，根据每个传感器期望其他传感器如何行动，最终的估计器甚至有可能比在没有任何战略意图的情况下更快地收敛到感兴趣的估计数量的正确值！换句话说，战略信息传输似乎对其目标系统既有好处，也有坏处。因此，工程师和系统设计师的中心问题是：“分界线在哪里？”这个项目涉及在受控系统中的虚假数据攻击、参与式传感应用中的故意错误报告以及对抗性机器学习的特定背景下的SIT的全面研究。它建立并扩展了经济学中所谓的“廉价谈话”文献中的现有模型，在该文献中，以前曾考虑过战略信息传输，尽管假设与网络-物理系统的安全有很大不同。这个项目的新颖性、针对性和智力优势在于：(1)它建立了新的模型，比现有框架更能解释利益三个应用中战略信息传递的特殊性，(2)它结合使用信息论和博弈论工具来分析这些模型，(3)使用行为经济学实验来帮助刻画和跨越实践中有害和有益的战略信息传递之间的边界。

As cyber-socio-physical and infrastructure systems are increasingly relying on data and integrating an ever-growing range of disparate, sometimes unconventional, and possibly untrusted data sources, there is a growing need to consider the problem of estimation in the presence of strategic and/or self-interested sensors. This class of problems, called "strategic information transmission" (SIT), differs from classical fault-tolerant estimation since the sensors are not merely failing or malfunctioning, but are actively trying to mislead the estimator for their own benefit. Such strategic behavior could happen through at least two mechanisms, sensor hijacking and willful misreporting, both of which have been observed in practice. One particularly stark and fundamental difference between strategic and faulty sensors is that a population of strategic sensors does not necessarily result in an erroneous estimate. In fact, depending on how each sensor expects others to act, it is even possible for the resulting estimator to converge to a correct value of the estimated quantity of interest faster than in the absence of any strategic intent! In other words, strategic information transmission appears to have the potential for being both a boon and a curse for the systems it targets. The central question for the engineer and system designer, then, is: "where is the demarcation line?"This project is concerned with a comprehensive study of SIT in the specific contexts of false data attacks in controlled systems, willful misreporting in participatory sensing applications, and adversarial machine learning. It builds on and extends existing models from the so-called "cheap talk" literature in Economics, where strategic information transmission has been considered before, although under widely different assumptions than in the context of security of cyber-physical systems. The novelty, pertinence, and intellectual merit of this project, lie in (1) its formulation of new models that more closely account for the specificity of strategic information transmission in the three applications of interest than existing frameworks, (2) its combined use of information theoretic and game theoretic tools to analyze these models and, (3) the use of behavioral economics experiments to help characterize, and straddle, the boundary between detrimental and beneficial strategic information transmission in practice.