基于补贴的异构多智能体系统稳定的博弈论方法
Game Theoretic Approach to the Stabilization of Heterogeneous Multiagent Systems Using Subsidy
摘要:
我们考虑一个由自私和异构代理组成的多智能体系统。它的行为是由多种群复制动力学模型,其中种群的支付函数是不同的。一般而言,复制子动态中存在多个平衡点。为了稳定一个理想的平衡点,我们引入了一个控制器,称为政府控制代理人的行为,提供补贴。在以前的工作中,假设政府根据代理人所属的人口来确定补贴。然而,一般来说,政府无法确定每个人口的成员。在本文中,我们假设政府观察每个代理人的行动,并根据观察到的行动配置文件确定补贴。然后,我们使用带有反馈的复制子动力学来建模代理的受控行为。我们推导了复制子动力学中目标平衡点的稳定条件。
英文摘要:
We consider a multiagent system consisting of selfish and heterogeneous agents. Its behavior is modeled by multipopulation replicator dynamics, where payoff functions of populations are different from each other. In general, there exist several equilibrium points in the replicator dynamics. In order to stabilize a desirable equilibrium point, we introduce a controller called a government which controls the behaviors of agents by offering them subsidies. In previous work, it is assumed that the government determines the subsidies based on the populations the agents belong to. In general, however, the government cannot identify the members of each population. In this paper, we assume that the government observes the action of each agent and determines the subsidies based on the observed action profile. Then, we model the controlled behaviors of the agents using replicator dynamics with feedback. We derive a stabilization condition of the target equilibrium point in the replicator dynamics.
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。