多智能体系统中的协作联盟：量化资源分配博弈的强无政府状态代价
Collaborative Coalitions in Multi-Agent Systems: Quantifying the Strong Price of Anarchy for Resource Allocation Games
AI总结：
本文将多智能体资源分配建模为共同利益博弈，以 $k$-强纳什均衡刻画协作联盟行为，并通过可处理线性规划给出 $k$-强无政府状态代价的紧界及效用可设计时的上界。
摘要：
新型通信技术的出现使我们能够扩展对分布式控制的理解，并考虑协作式决策范式。借助协作算法，某些局部决策实体（或智能体）能够相互通信并在行动上开展协作，以获得更好的系统行为。通过限制通信量，这些算法处于集中式方法与完全分布式方法之间。为了理解这种智能体间协作可能带来的收益，我们将多智能体系统建模为一种共同利益博弈，其中多组智能体可以在行动上协作，以共同提高系统福利。我们特别考虑将 $k$-强纳什均衡作为这些系统的涌现行为，并探讨这些状态对系统最优解的近似程度，该程度由 $k$-强无政府状态代价比来形式化。我们的主要贡献在于，给出有限资源分配博弈中 $k$-强无政府状态代价的紧界，并将其作为一个可处理线性规划的解。通过改变 $k$——即协作联盟的最大规模——我们可以准确观察智能体间协作能够带来多少性能提升。为了研究进一步的改进机会，我们还在智能体效用函数可被设计的情况下，给出可达到的最大 $k$-强无政府状态代价的上界。
英文摘要：
The emergence of new communication technologies allows us to expand our understanding of distributed control and consider collaborative decision-making paradigms. With collaborative algorithms, certain local decision-making entities (or agents) are enabled to communicate and collaborate on their actions with one another to attain better system behavior. By limiting the amount of communication, these algorithms exist somewhere between centralized and fully distributed approaches. To understand the possible benefits of this inter-agent collaboration, we model a multi-agent system as a common-interest game in which groups of agents can collaborate on their actions to jointly increase the system welfare. We specifically consider $k$-strong Nash equilibria as the emergent behavior of these systems and address how well these states approximate the system optimal, formalized by the $k$-strong price of anarchy ratio. Our main contributions are in generating tight bounds on the $k$-strong price of anarchy in finite resource allocation games as the solution to a tractable linear program. By varying $k$ --the maximum size of a collaborative coalition--we observe exactly how much performance is gained from inter-agent collaboration. To investigate further opportunities for improvement, we generate upper bounds on the maximum attainable $k$-strong price of anarchy when the agents' utility function can be designed.
机构、总结、中文摘要均由AI分析而来，可能存在错误，请以论文正文为准。
