Strategic Interactions in Multiagent Systems

自利的或竞争性的多代理系统是指多个代理交互，但可能具有不同的目标和信息的设置。这些系统中的一些关键问题是：1)智能体应该如何以及何时协调它们的行动和决策；2)如何设计协议来处理智能体的自身利益并实现系统范围内的理想目标。自利多智能体系统的例子比比皆是，从投票到游戏，多智能体系统研究社区的想法对从港口安全到智能电网的应用产生了重大影响。在许多多智能体系统中，智能体之间的相互作用在某种程度上受到约束。例如，潜在的社会网络可能鼓励和支持某些相互作用，或者运输网络可能促进某些代理之间的资源交换，同时在其他代理之间形成障碍。因此，在研究和讨论战略行为时，重要的是要考虑和理解潜在的约束如何影响代理的决策和行动，以及这如何导致不同的结果。我将研究多智能体系统，其中智能体相互作用的约束可以通过图或网络显式建模。我把这样的多智能体系统简称为“网络化多智能体系统”或“网络化MAS”。我将为这些系统开发适当的模型，并研究由此产生的博弈论和计算问题。这项工作的意义在于加深了我们对自利益主体如何组织和协调其行为的理解，这是多智能体系统和人工智能的一个关键研究问题。这项研究的结果还将在电子商务、社交网络和智能电网等领域带来新的应用。

Self-interested, or competitive, multiagent systems are settings where multiple agents interact while potentially having different goals and information. Some of the key issues in these systems are i) how and when should agents coordinate their actions and decisions and ii) how to design protocols to handle the agents' self-interest and achieve system-wide desirable goals. Examples of self-interested multiagent systems abound, ranging from voting to game-playing, and ideas from the multiagent systems research community have had significant impact in applications ranging from port security to the Smart Grid.In many multiagent systems the interactions of the agents are constrained in some way. For example, the underlying social network may encourage and support certain interactions over others, or the transportation network may facilitate resource exchange between certain agents while forming a barrier between others. Thus, when studying and discussing strategic behaviour it is important to account for and understand how the underlying constraints affect the decisions and actions of the agents and how this leads to different outcomes.I will investigate multiagent systems where the constraints on agents' interactions can be explicitly modelled by a graph or network. I call such multiagent systems "networked multiagent systems" or "networked MAS" for short. I will develop appropriate models for these systems and investigate the game-theoretic and computational issues that arise. The significance of the work lies in deepening our understanding of how self-interested agents should organize and coordinate their actions, a key research problem in multiagent systems and artificial intelligence. Results from this research will also lead to novel applications in areas such as electronic commerce, social networks and the Smart Grid.