Abstract
Small-world and scale-free properties are important features of networks in the real world, which will make learning in such networks difficult. One of the defining characteristics of small-world networks is a high cluster coefficient. In the present study, processes of reinforcement learning in complex networks with different clustering structures are investigated. From the results of the experiments, clusters in the networks of learning spaces could be learned in more trials with relatively high discount rates and/or rewards. Thus, clusters in the learning spaces may destabilize a process of reinforcement learning within a certain range of parameter settings. It is discussed that solutions affected by the clusters are less optimal, but they may be more robust.
