CAREER: Cooperative Multi-Agent Optimization

这一学院早期职业发展(CALEAR)奖为研究和教育活动提供资金，这是一个共同的优化主题。研究目标是为大规模分布式多智能体系统建立新的计算模型、理论进展和优化算法。令人感兴趣的是由具有不同性能标准的互连的多个代理组成的系统。由于各种原因，如私人或专有信息，代理不共享他们自己的目标，但确实共享稀缺的资源，并希望合作实现共同目标。在没有中央协调器或中央信息访问的情况下，这种多智能体系统的协调和优化必须是分布式的。一个主要的研究目标是开发和研究数学模型，并设计和分析分布式多代理算法。在分布式模型中，当代理连通性随时间动态变化时，每个代理在本地行动，并与其邻居共享一些有限的信息。另一个目标是探索和量化算法在各种系统特征下的性能极限，例如通信噪声或延迟的存在。算法的发展需要一些基础研究，为系统性能的分析和表征提供新的数学工具。这项研究紧密结合教育计划，开设新的本科和研究生优化课程，使学生具备有效、系统地识别、建模、分析和解决优化问题的知识。更广泛的教育目标包括促进妇女和其他代表不足的群体对优化的兴趣，以及向外宣传和教育年轻人关于优化的重要性和美感。研究活动的成功完成将导致为大型网络系统设计新的高效的分散协调和优化算法。此外，它还将导致针对一大类非线性非凸问题设计出性能有保证的全局优化算法。总体而言，结果将从总体上增强现有的优化知识。计划中的教育活动将促进学生群体的优化和多样性。

This Faculty Early Career Development (CAREER) award provides funds for research and education activities on a common theme of optimization. The research objective is to establish new computational models, theoretical advances, and optimization algorithms for large scale distributed multi-agent systems. Of interest are systems that consist of interconnected multiple agents with different performance criteria. For various reasons, such as private or proprietary information, the agents do not share their own objectives, but do share scarce resources and want to cooperatively achieve a common goal. In the absence of a central coordinator or central information access, the coordination and optimization of such multi-agent systems have to be distributed. A primary research objective is to develop and study mathematical models, and design and analyze distributed multi-agent algorithms. In the distributed model, each agent acts locally and shares some limited information with its neighbors while the agent connectivity is dynamically changing with time. Another objective is to explore and quantify the performance limits of the algorithms under various characteristics of the system, such as the presence of communication noise or delays. The algorithmic development necessitates some fundamental research providing new mathematical tools for analysis and characterization of the system performance. The research is closely tied with educational plans to build new undergraduate and graduate optimization courses to equip the students with the knowledge to recognize, model, analyze, and solve optimization problems efficiently and systematically. A broader educational goal includes promoting interest for women and other under-represented groups in optimization, as well as outreaching and educating young minds about the significance and beauty of optimization.Successful completion of the research activities will lead to new efficient designs of decentralized coordination and optimization algorithms for large network systems. Also, it will lead to the designs of global optimization algorithms with guaranteed performance for a large class of non-linear non-convex problems. Overall, the results will enhance the existing knowledge in optimization in general. The planned educational activities will promote optimization and enhance the diversity in the student population.