Keeping Secrets: Realizing the Potential of Decentralized Discrete-Event Systems

点击翻译按钮获取中文摘要

The proposed research program will develop decentralized discrete-event systems (DES) to support autonomous vehicles (AV), smart cities, power grids and other interconnected networks of communicating agents by ensuring security and opacity. Power and electricity generation is becoming decentralized and self-driving cars are by nature decentralized. Decentralization requires the communication of information (such as event occurrences) between disparate agents (e.g., AVs, power generators, or computers), which opens a system to possible security breaches or hostile maneuvering. The proposed program will develop new techniques for maintaining system opacity; in DES, opacity means that sequences of events or states that have been designated as secret cannot be distinguished from other, non-secret event sequences or states. If a system is opaque, hostile agents will not have information critical to mounting an attack on the system. For agents that are distributed across a system to work together to ensure security, information or knowledge that one agent lacks must be supplied (in this case, communicated) by another agent with different, and complementary, information. While prior work in DES has examined communication between agents to enact control, existing research has not focused on communication to ensure opacity. The problem with distributed communication is that agents must find ways to help other agents perform tasks while ensuring that an adversary does not learn enough about the system to discern secret information. Deriving communication strategies requires understanding what information each agent possesses. Just as people make decisions based not only on their direct observations but also on what they know about what other people know, so too should automated agents make decisions based on their knowledge of other agents. Consequently, the proposed research will also develop a mathematical-logic model for DES that captures the knowledge that agents have about the system and about other agents. Mathematical models of knowledge have been used, to advantage, to model problems in computer science but have yet to be significantly exploited in control systems or DES. As an example, to make "smart cars" truly smart we need to imbue these unmanned vehicles with the capability to reason about the knowledge that other vehicles have so that vehicles can independently determine what information to send and when to send it in a way that ensures that the system is impervious to being commandeered by unauthorized agents. My research program will train 4 PhD and 3 MASc students in DES modeling, algorithm design, and software development and will enable HQP to test and refine their research on real networks of interconnected computers and autonomous vehicles. My program will enable Canada to meet societal demands for ensuring that as we move to more distributed, more intelligent and more autonomous systems, we do not sacrifice security and safety.