Effects of adaptive agent reinforcement learning on cooperation in spatial networks

In light of the dynamic and heterogeneous nature of individual preferences in real-world social interactions, we propose a research framework that integrates reinforcement learning with adaptive weights. In real social systems, interactions among individuals are not heterogeneous, and they exhibit significant preference diversity and context dependence. By introducing dynamic weight evolution, this study characterizes the adaptive processes of individuals in social dilemmas, where weight allocation captures the dual influence of social interaction preferences and feedback from stimuli. The results reveal that this adaptive weighting mechanism effectively sustains high levels of cooperation within the population. In particular, adaptive weights stimulate cooperation through preference selection under high social dilemma strength. Importantly, the evolution of weights leads to a polarization of the strategy within the population, resulting in a stable coexistence of cooperation and defection. Furthermore, the study uncovers the micro-level regulatory role of the weight mechanism in network reciprocity, enriching the theoretical framework of cooperation evolution in complex systems. It provides a new perspective for understanding the formation and evolution of cooperation.