Trustworthy Autonomous Intelligent Systems

现代社会依赖于各种系统和基础设施提供的服务和应用：数据中心、跨越核心、边缘和传感器/物联网计算的云、互联网服务提供商网络、蜂窝网络、多式联运网络和电网。这些服务和应用基础设施的特点越来越多样化，它们的要求要求更复杂、更精确和更灵敏的控制。例如，电网不再只提供单向的能源输送，现在必须支持不同类型的可再生能源和传统能源的动态电力转移，以满足现有的和新的电力需求，如电动汽车。此外，在移动性方面，从私家车过渡到无缝的多模式出行(火车、公交车、地铁、共享车辆和自行车，以及步行)。此外，移动性、能源、计算和网络基础设施必须考虑碳足迹和空气污染等环境影响。 该研究计划的目标是促进对值得信赖的自主智能控制和管理系统(CMS)的理解、创建、设计和操作，CMS可指导大规模基础设施中的资源分配，以实现在质量、数量/数量和成本方面不同的服务和应用程序的交付。CMSS在需求服务和资源可用性之间进行调解。其中许多服务是关键任务，因此在供需波动、故障和攻击存在的情况下，CMS是值得信赖和可靠的。基础设施的规模和复杂性要求CMS利用人工智能(AI)和机器学习(ML)来实现自主操作。为了实现我们的CMS目标，我们应对四个挑战： 1.意向驱动的网络。如何将有关政策和意图的高级别声明转化为高保真的低级别管理和控制行动？ 2.自主操作的分析与学习。如何将分析、AI和ML应用于分布式大型基础设施的管理和控制，以实现值得信赖的自我优化、自我配置、自我修复、自我保护的自适应行为？ 3.机制设计：激励与定价。如何将高水平的全球目标转化为激励和定价信号，从而引导全球受益的用户资源消耗？ 我们的CMSS方法包括：1.算法设计；2.软件实现；3.概念验证系统。 我们的研究影响将是： 在计算云、物联网和5G网络中安全、专用、响应迅速且可靠地交付应用 通过多模式出行、自动驾驶车辆和按需移动，大幅减少道路上的车辆数量 通过可再生能源和汽车电气化缓解和消除大都市的污染。

Modern society relies on services and applications that are delivered by a variety of systems and infrastructures: data centers, clouds spanning core, edge, and sensor/IoT computing, Internet service provider networks, cellular networks, multimodal transportation networks, and power grids. These services and applications infrastructures are increasingly more diverse in their characteristics and their requirements call for more sophisticated, precise and responsive control. For example, power grids no longer provide only unidirectional delivery of energy, but must now support dynamic power transfer from different types of renewable and conventional energy sources to meet existing and new power demands, such as electric vehicles. As well, in mobility there is a transition from private cars to the seamless multimode travel (trains, buses, subways, shared vehicles and bicycles, and walking). Furthermore, mobility, energy, computing and networking infrastructures must consider environmental impacts such as carbon footprint and air pollution. The goal of this research program is to advance the understanding, creation, design, and operation of trustworthy autonomous intelligent control and management systems (CMSs) that direct the allocation of resources in large-scale infrastructures to enable the delivery of services and applications that vary in terms of quality, quantity/volume, and cost. The CMSs mediate between the demand services and availability of resources. Many of these services are mission critical, and so the CMS be trustworthy and reliable, in the presence of fluctuations in supply and demand, and of failures and attacks. The scale and complexity of infrastructures require that the CMS leverage artificial intelligence (AI) and machine learning (ML) to enable autonomous operation. We address four challenges to achieve our CMS goal: 1. Intent-Driven Networking. How can high-level statements regarding policies and intent be translated into high-fidelity low-level management and control actions? 2. Analytics and Learning for Autonomous Operation. How can analytics, AI and ML be applied in management and control of distributed large-scale infrastructures to enable trustworthy self-optimizing, self-configuring, self-healing, self-protecting adaptive behavior? 3. Mechanism Design: Incentives and Pricing. How can high-level global objectives be translated into incentives and pricing signals that induce globally beneficial user resource consumption? Our methodology for CMSs integrates: 1. Algorithm design; 2. Implementation in software; 3. Proof-of-concept systems. Our research impacts will be: Secure, private, responsive and reliable delivery of applications in computing clouds, IoT and 5G networks Dramatic reduction in the number of vehicles on the road through multimode travel, autonomous vehicles and Mobility-on-Demand Mitigation and elimination of pollution in metropolises through renewable energy and vehicle electrification.