Machine against machine: towards a better understanding of future cyber physical system security

该项目旨在解决保护网络物理系统免受复杂网络攻击的问题。网络物理系统(CP)将计算机和通信技术与感知物理世界并与之交互的设备相结合，以实现所需的物理效果。它们涵盖广泛的系统，包括工业控制系统(ICS)、关键基础设施、自动驾驶车辆和消费物联网(IoT)设备。这些系统由集中式或自主控制算法控制，这些算法使系统的物理部分以所需的方式运行。这一提议解决了更复杂的网络攻击带来的问题，这些攻击正是针对这些控制算法，向它们提供虚假信息或危及系统的关键组件。由于攻击很可能是由自动化过程进行的，针对它们部署的防御机制也必须是自主的。这是一种与传统信息技术(IT)安全截然不同的情况，在传统信息技术(IT)安全中，防御过程大多是由人驱动的，需要新的理论框架来描述这些机器对机器的情况将如何随着时间的推移而演变，因为防御和进攻过程相互优化，同时由于它们试图防御和攻击的网络物理系统的操作环境发生重大变化而受到进化压力。拟议项目的目标有两个。首先，我们将尝试建立基于混合人工智能(AI)的防御机制，该机制结合了基于规则的推理专家系统和更新的机器学习方法的优势。为此，我们将使用本体，一个成熟的知识表示工具，以构建高级概念模型的网络和物理组件的CP的启发，他们试图控制的物理世界的法律和规则。为了测试和评估这种防御过程，我们将扩展我们为一些特定应用领域(如电子网络和空中交通管制系统)开发的实验性CPS仿真方法，并将其应用于新的领域，以验证使用本体的灵活性和一般可行性。其次，我们试图通过运用博弈论和数学最优化的概念和数学构造，建立一个理论框架来描述CPS背景下的机器对机器军备竞赛中的协同优化和协同进化现象，以便我们能够更好地理解、建模和预测这种军备竞赛的最终结果。

This project proposes to address the problem of protecting Cyber Physical Systems against sophisticated cyber attacks. Cyber physical systems (CPS) combine computer and communications technology with devices that sense and interact with the physical world in order to achieve a desired physical effect. They encompass a wide array of systems including Industrial Control Systems (ICS), critical infrastructure, autonomous vehicles, and consumer Internet of Things (IoT) devices. These systems are controlled by centralized or autonomous control algorithms that make the physical portions of the system behave in a desired manner. This proposal addresses the problem posed by the more sophisticated cyber attacks that target precisely these control algorithms by providing them with false information or compromising key components of the system. Since attacks will most likely be conducted by automated processes, the defense mechanisms deployed against them will also have to be autonomous. This is a very different situation from traditional Information Technology (IT) security where defensive processes are mostly human-driven, and required new theoretical frameworks to describe how these machine-against-machine situations will evolve over time, as both defensive and offensive processes optimize against each other and are simultaneously subject to evolutionary pressures due to significant changes in the operational environments of the cyber physical systems they are trying to defend and attack. The goal of the proposed project is two-fold. First, we will attempt to build hybrid Artificial Intelligence (AI) based defensive mechanisms that combine the advantages of rule-based reasoning expert systems and more recent machine-learning approaches. To do so, we will use ontologies, a well-established knowledge representation tool, to construct high level conceptual models of the cyber and physical components of CPS inspired by the laws and rules of the physical world they are trying to control. To test and evaluate such defensive process we will expand on the experimental CPS emulation approach we have developed for a few specific application domains, such as electrical networks and Air Traffic Control systems, and apply to new domains, in order to validate the flexibility and general viability of the of use of ontologies. Second, we will attempt to develop a theoretical framework for describing the phenomena of co-optimisation and co-evolution in the machine-against-machine arms-race in the context of CPS, by applying concepts and mathematical constructs from Game Theory and mathematical optimization, so that we can better understand, model and predict the eventual outcome of such arms races.