Decentralized optimization of dynamic multi-agent networks

点击翻译按钮获取中文摘要

What if we could guarantee some prescribed global behavior of complex, dynamically evolving networks by devising simple rules that locally guide the behavior of individual network nodes? Systems such as communication networks, power grids, transportation systems are all examples of networks comprising many sub-systems that often interact in dynamic ways. Such networks can be regarded as multi-agent networks, i.e., complex, networked systems composed of many relatively simple components, or "agents", that sense only their local environment, and have the ability to affect it, and the other agents within their proximity. Centralized control of such networked systems impossible, yet our need to guarantee their robust operation is becoming vital. Our goal is to develop a theory towards their decentralized optimization and control. Such a theory does not exist yet, it is a long-time out, but it is achievable. Game theory and optimization are both powerful frameworks within which design can be done. Regardless of the specific application domain, the goal is to design local control policies for the individual agents or sub-systems to ensure that the collective behavior is desirable with respect to the system objective. The long-term goal is to have a theory for decentralized optimization of dynamic multi-agent networked systems. Because of its power of abstraction such research has applicability to diverse fields and the potential to be transformative.