CIF: Small: Communication-Aware Decentralized Game-Theoretic Learning Algorithms for Networked Systems with Uncertainty

许多新兴技术联网系统(如物联网、智能电网、交通系统)机会的核心在于去中心化，这意味着部署相互关联的代理，例如传感器、家用电器和设备，它们以协调的方式运行。在这些系统的分散运作中，一个隐含的假设是，这些代理人同意或能够解决他们在目标上的分歧。在大规模网络系统中，除非环境信息的不合理准确性和不合理的协调水平，否则代理无法确定其他代理正在优化什么。该项目将为技术联网系统中的信号处理和通信问题提供实用和可扩展的分散解决方案，在这些系统中，无线网络技术为信息交换提供骨干。此外，该项目将包括指导博士生，为本科生提供研究经验，以及在德克萨斯农工大学开发一门关于学习和网络的研究生课程。该项目的前提是，在网络多智能体系统中，当智能体由于环境的不确定性而对自己的目标产生分歧时，他们是在与不确定性进行博弈，然后用博弈论的概念来定义最优性准则。在采用博弈论均衡作为最优标准后，该项目被分为三个主要推力。第一个推力将开发基于博弈论概念的分散算法，并将表征当大规模网络系统中的信息聚合或共识可能无法实现时的有限时间结果。第二个重点是了解通信与最优之间的基本权衡，并开发导致通信成本高昂和错误的去中心化算法。第三个推动力将寻求描述收敛到不受欢迎的均衡的可能性，并将使用影响力最大化的方法来避免糟糕的结果。总体而言，该项目借鉴了分散优化中常见的方法，如共识，以及信号处理中的常见方法，如通信审查，并将它们与博弈论、最佳响应型、学习算法相结合，以设计可扩展和具有通信意识的分散算法。该奖项反映了NSF的法定使命，并通过使用基金会的智力优势和更广泛的影响审查标准进行评估，被认为值得支持。

At the core of the opportunities for many emerging technological networked systems, such as Internet-of-things, smart grid, transportation systems, lies decentralization which means the deployment of interlinked agents, e.g., sensors, appliances, and devices, that operate in harmony. An implicit assumption in the decentralized operations in these systems is that these agents agree or are able to resolve their differences in what is their objective. In large scale networked systems, barring unreasonable accuracy of environmental information and unjustifiable levels of coordination, agents cannot be sure of what other agents are optimizing. This project will enable practical and scalable decentralized solutions to signal processing and communication problems in technological networked systems where wireless network technologies provide the backbone for information exchange. Furthermore, this project will involve mentoring of PhD students, providing research experience to undergraduate students, and the development of a graduate-level course on learning and networks at Texas A&M University. The premise of this project is that when agents disagree on their objectives in a networked multi-agent system due to environmental uncertainty, they are playing a game against uncertainty, and the optimality criterion is then defined by game-theoretic notions. Upon adopting, game-theoretic equilibria as the optimality criterion, the project is divided into three major thrusts. The first thrust will develop decentralized algorithms based on game-theoretic notions, and will characterize finite-time outcomes when information aggregation or consensus in large-scale networked systems may not be possible. The second thrust aims to understand fundamental trade-offs in communication vs. optimality, and to develop decentralized algorithms that account for costly and faulty communication. The third thrust will seek to characterize the likelihood of converging to undesirable equilibria, and will use influence maximization methods to avoid bad outcomes. Overall, the project draws upon methods common in decentralized optimization, e.g., consensus, and in signal processing, e.g., communication censoring, and combines them with game-theoretic, best-response type, learning algorithms to design scalable and communication-aware decentralized algorithms.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.