CAREER: Directed Information Theory for Networked Control Systems in Big Data Regime

支持当今关键基础设施（例如，交通网络，电力系统）的现代控制系统由大量通过通信网络互连的计算设备组成。这样的系统通常被称为网络控制系统（NCSs），在过去的二十年中，从理论和实践的角度来看，一直是研究兴趣的主题。无线和有线通信技术的成熟使各种NCS概念成为可能（例如，联网汽车，远程手术，基于云的控制），NCS的作用将在未来进一步发展。然而，最近，人工智能（AI）技术的进步正在增加对基于NCS架构的AI驱动控制算法实现的需求。虽然这些新趋势体现了NCS的卓越能力，但它们在以下方面对当前形式的NCS技术提出了前所未有的挑战：(1)整合支持新兴人工智能控制方案的大数据内容的能力，(2)在有限的网络资源内容纳大量参与者的可扩展性，以及(3)针对潜在对手的安全性和隐私性。本CAREER提案的目的是应用一种新的信息流优化技术，称为最优信息流（OIF）合成，以解决上述三个领域中几个选定但具有代表性的技术挑战。提出的方法的关键促成因素和独特性是定向信息的新使用，定向信息是一种信息论数量，允许在网络上识别和确定任务相关信息流的优先级。我们的出发点是零延迟率失真理论及其在单通道ncs中的成功应用。为了解决(1)，我们应用定向信息为大容量数据流开发了一种实时的、任务相关的数据压缩算法。我们的目标应用包括用于自动驾驶的基于云的视觉伺服，我们研究在不降低控制性能的情况下，视频流数据速率降低的可能性。为了解决(2)，我们通过将现有的NCS理论与网络信息理论相结合，寻求多通道NCS设计的新方法。我们将展示适当的控制-通信协同设计可有效降低自动驾驶车辆排的燃油消耗。对于(3)，我们将研究基于云的控制系统的加密和非加密隐私保护机制。我们提出了基于OIF综合实现最优效用-隐私平衡的设计原则。我们不是以特定领域的方式来执行这三个研究任务，而是强调定向信息在每个任务中的潜在作用，从而有助于将控制理论与信息论相结合的公认的学术挑战。该CAREER提案还涉及创新的教育活动，包括开发独特的ncs研究生课程，以最大限度地发挥研究和教育之间的协同效应。该奖项反映了美国国家科学基金会的法定使命，并通过使用基金会的知识价值和更广泛的影响审查标准进行评估，被认为值得支持。

Modern control systems supporting today’s critical infrastructure (e.g., transportation network, power systems) are comprised of a large number of computational devices interconnected over communication networks. Such systems are commonly called Networked Control Systems (NCSs), and have been the subject of research interest over the last two decades, both from theoretical and practical perspectives. The maturation of technologies in both wireless and wired communication made various NCS concepts realizable (e.g., connected cars, remote surgery, cloud-based control), and the roles of NCSs will further grow in the future. Recently, however, the advancements of Artificial Intelligence (AI) technologies are increasing demand for AI-powered control algorithm implementations over NCS architectures. While these new trends exemplify the remarkable capabilities of NCSs, they raise unprecedented challenges to the current form of NCS technologies in terms of: (1) capacity to incorporate Big Data contents that support emerging AI-based control schemes, (2) scalability to accommodate a large number of players within limited network resources, and (3) security against and privacy from potential adversaries. The purpose of this CAREER proposal is to apply a novel information flow optimization technique, termed Optimal Information Flow (OIF) synthesis, to address a few selected, yet representative, technological challenges in each of the three areas identified above. The key enabler, as well as the uniqueness, of the proposed approach is the novel use of directed information, an information-theoretic quantity that allows for the identification and prioritization of task-relevant information flow over the network. Our starting point is the zero-delay rate-distortion theory and its successful applications to single-channel NCSs. To address (1), we apply directed information to develop a real-time, task-dependent data compression algorithm for high-volume data streams. Our target applications include cloud-based visual servoing for autonomous driving, where we study the extent that data-rate reduction from video streams is possible without deteriorating control performance. To tackle (2), we pursue new approaches to multi-channel NCS design by blending the existing NCS theory with network information theory. We will demonstrate that an appropriate control-communication co-design effectively reduces fuel consumption of autonomous vehicle platoons. For (3), we will study both cryptographic and non-cryptographic privacy preserving mechanisms for cloud-based control systems. We propose design principles for achieving the optimal utility-privacy balance based on the OIF synthesis. Instead of carrying out these three research tasks in a domain-specific manner, we emphasize the underlying roles of directed information in each task, thereby contributing to the well-recognized academic challenge of integrating control theory with information theory. This CAREER proposal also involves innovative educational activities, including the development of a unique graduate-level curriculum on NCSs, in order to maximize the synergistic effects between research and education.This award reflects NSF's statutory mission and has been deemed worthy of support through evaluation using the Foundation's intellectual merit and broader impacts review criteria.