网络演化博弈实验中的策略学习动力学与相变问题研究

项目名称： 网络演化博弈实验中的策略学习动力学与相变问题研究

项目编号： No.11475074

项目类型： 面上项目

立项/批准年度： 2015

项目学科： 数理科学和化学

项目作者： 关剑月

作者单位： 兰州大学

项目金额： 80万元

中文摘要： 近几年来的一些真人博弈实验结果表明:以往博弈理论模型中所提出的个体策略更新动力学都不能准确地描述群体层面的策略演化以及系统合作行为的涌现现象。在本项目中，我们将开展大规模真人网络博弈实验，通过详细跟踪个体所采取策略与所获得收益的时间序列，分析其与网络邻居的策略与收益时间序列的关联行为，探寻影响个体决策动力学的关键性因素。同时根据行为心理学的研究成果，研究不同的强化学习规则对群体合作行为演化的影响，通过调节学习速率参数，找到理论预测结果与真人实验结果最符合的规则，为更好地理解实际社会系统中合作演化提供科学指导。最后我们还将借用统计物理学和非线性动力学中的思想和方法来详细研究网络博弈演化过程中呈现出来的相变行为，探究影响其临界指数或普适类的结构因素和动力学机制，为深入理解复杂适应系统局部微观作用导致宏观涌现行为提供深刻的理论解释。

中文关键词： 演化博弈；复杂网络；策略学习动力学；复杂性科学；统计物理

英文摘要： The comprehensive results of recent human game experiments indicate that the strategy-updating rules proposed in previous game theoretical models cannot characterize how our humans change strategies and how the collective cooperation emerge in the real social population. In this project, we plan to implement large-scaled networked game experiments with participators recruited from our universities. We expect to find the crucial factors affecting ones strategy decision-making processes, by trailing the time series of each person's strategy and payoff information and by investigating how these two series correlate with those of the neighbors. Meanwhile, inspired by the results from behavioral psychology field, we want to study how the various reinforcement learning rules influence the evolution of collective cooperation. By tuning the parameters of reinforcement learning rules, we expect to find the optimal ones that can better match those results from real experiments to guide our understanding how cooperation evolves in real social systems. Besides, we will also explore the phase transition phenomena arisen in networked games by using the methods and ideas from statistical physics and nonlinear dynamics. By analyzing the critical exponents, universality classs, and their dependence on the network topology and dynamical rules, our results will hope to help us deep understanding how local interactions lead to global emergent phenomenon in complex adaptive systems.

英文关键词： evolutionary games;complex network;strategy-learning dynamics;complexity science;statistical physics