Agent之间重复交互策略研究

在Agent构成的社会中信息可以共享,由于单一的交互对共享信息是没有作用的,因此,提出重复的交互策略。重复交互可以使Agent彼此共享信息,通过惩罚来达到系统的平衡,惩罚是通过忽视被惩罚的Agent的询问来实现的。通过简单地介绍一个周期模型,并且定义了延迟的折扣因子和预期效用,从而考虑了更一般的情况,即建立了n个周期模型,描述了通过观察询问的历史记录中没有应答的情况,来决定是否执行惩罚,并且通过试验进行了验证。随着n的增加,Agent的预期效用增加,而平衡的存在越来越少。

Research of Repeated Interaction Strategies Among Agents

Information can be shared in the multi-agent's society.A single interaction is often nonbeneficial,so repeated interaction strategy is given.Repeated interaction can make information sharing,and it can get equipoise through punishment by which ignoring its enquiry do.In the paper,there is a simple period model,and discount gene and expectation avail are defined.So,designs n-period model.Punishment is done by watching its history that did not answer.It has been validated by an experimentation.Equipoise exists less with n and expectation avail is increasing.