Scholarlink

<p>Modeling network evolution is foundational for understanding and regulating networks. Existing models simplify away the fact that network evolution is often a group decision process, leading to two primary limitations: that nodes lack the ability to learn policy and that there is no coordination among node policies. To address these shortcomings and consider the effectiveness of multi-agent reinforcement learning in solving group decision tasks, this paper proposes a complex Network Evolution model based on Multi-Agent Reinforcement Learning (NEMARL). In our model, swarm intelligence emerging through collaborative interactions among autonomous nodes drives the evolution of the network structure. Our extensive experiments demonstrate that the NEMARL model accurately reproduces classical network characteristics and fits real network data well. Furthermore, we demonstrate its effectiveness through scenario testing.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Modeling network evolution by multi-agent reinforcement learning

  • 期刊论文

  • 出版模式：

    开放获取

  • 发表日期：

    2026年7月21日

查看全文
  • Dong Li,
  • Tianwei Lin,
  • Zhaoyang Bao,
  • Bingqiao Gu,
  • Yatao Zhang,
  • Fei Yang,
  • Yanhao Sun,
  • Zhanwei Du,
  • Petter Holme

摘要

Modeling network evolution is foundational for understanding and regulating networks. Existing models simplify away the fact that network evolution is often a group decision process, leading to two primary limitations: that nodes lack the ability to learn policy and that there is no coordination among node policies. To address these shortcomings and consider the effectiveness of multi-agent reinforcement learning in solving group decision tasks, this paper proposes a complex Network Evolution model based on Multi-Agent Reinforcement Learning (NEMARL). In our model, swarm intelligence emerging through collaborative interactions among autonomous nodes drives the evolution of the network structure. Our extensive experiments demonstrate that the NEMARL model accurately reproduces classical network characteristics and fits real network data well. Furthermore, we demonstrate its effectiveness through scenario testing.