複雑ネットワークにおける経路学習問題に関する研究

概要

論文の詳細を見る
Many networks in the real world show the scale-free property in which degree distribution of a network follows a power-law. It has been suggested that scale-free networks are vulnerable to attack on the highly connected nodes, called hubs. Thus, path learning in scale-free networks faces problems of optimality and attack-tolerance of learned path. In this study, I attempted to obtain attack-tolerant behavior of the reinforcement learning agent by employing the cost function and tuning the three learning parameters. The cost function aiming at avoiding paths consisting of hubs improved the attack-tolerance. Tuning of a parameter called the discount rate was also found to be effective in decreasing the attack damage.
滋賀大学教育学部の論文