非合作环境下的决策方法研究

在对理性决策者决策过程建模基础上,研究了信息不对称的非合作环境中,存在着决策者效用函数U的先验概率分布p(u)(比如从一组效用函数中学习)的条件下,通过对决策者所已采取的行动的观察,生成效用函数U的约束条件集合,通过这些约束条件来确定其效用函数U后验分布q(u),从而推断他的效用函数U,用以预测他未来的行动的方法。本文提出了从策略观察中提取约束条件的算法,以及在非合作环境中,具有学习能力的谋略型决策者的行动策略。通过一个实例,研究相关的算法并进行了相应的实验。

Research on Decision-making In Non-cooperative Settings

Based on modeling of rational decision maker,The paper introduces a new mean of decision-making in information asymmetric non-cooperative setting. With the assumption of the prior distribution p(u) over decision-maker'utility function U known,we observe a decision-maker previous actions to obtain a constrain set of U. We can condition on these constraints to obtain a posterior distribution q(u) to predict the decision-maker possible actions. We show that this capability is particularly useful in a non-cooperative setting. The paper introduces the main algorithm and outcome of experiment.