Decentralized optimization of dynamic multi-agent networks

如果我们可以通过设计简单的规则来局部指导单个网络节点的行为，从而保证复杂的、动态发展的网络的某些规定的全局行为，那会怎么样？通信网络、电网、交通系统等系统都是由许多子系统组成的网络，这些子系统经常以动态的方式相互作用。这样的网络可以被看作是多智能体网络，即由许多相对简单的组件或“智能体”组成的复杂的网络系统，这些“智能体”只感知其局部环境，并有能力影响它，以及它们附近的其他智能体。集中控制这样的网络系统是不可能的，但我们需要保证他们的稳健运行变得至关重要。我们的目标是发展一种理论来实现它们的分散优化和控制。这样的理论目前还不存在，它是一个漫长的过程，但它是可以实现的。博弈论和优化都是可以进行设计的强大框架。无论具体的应用领域是什么，目标都是为单个代理或子系统设计本地控制策略，以确保集体行为符合系统目标。长期目标是为动态多智能体网络系统的分散优化提供理论依据。由于其抽象的力量，这种研究适用于不同的领域，并具有变革性的潜力。

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.