Scholarlink

The rapid development of Large Language Models (LLMs) is transforming artificial intelligence, leading to the emergence of powerful collaborative learning frameworks such as federated learning. However, ensuring secure and reliable intelligence sharing among distributed LLMs remains a critical challenge due to dynamic network conditions and heterogeneous data sources. This paper proposes a Dynamic Incentive Sharing Mechanism (DISM) to encourage collaborative intelligence sharing among LLMs in blockchain-based federated learning systems. By providing tamper-proof records of interactions and rewards, blockchain ensures the security and integrity of the learning process. DISM leverages evolutionary game theory to promote cooperation and address challenges such as data poisoning, mitigating disparities in shared intelligence. By calculating each node’s contribution and distributing rewards accordingly, DISM fosters active participation and enhances the overall performance of federated LLM fine-tuning. Experimental simulations and comparative analyses validate the effectiveness of DISM, demonstrating its potential to improve the reliability and efficiency of blockchain-enabled collaborative intelligence sharing.

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

Incentive Mechanisms for Collaborative Intelligence Sharing in Blockchain-Based Federated LLM Fine-Tuning

  • 会议论文

  • 发表日期：

    2026年

查看全文
  • Jinyu Zhang,
  • Yixuan Pan,
  • Zhaomin Wu,
  • Rongqi Zhou,
  • Yi Yang,
  • Peng Wang,
  • Wen Sun

摘要

The rapid development of Large Language Models (LLMs) is transforming artificial intelligence, leading to the emergence of powerful collaborative learning frameworks such as federated learning. However, ensuring secure and reliable intelligence sharing among distributed LLMs remains a critical challenge due to dynamic network conditions and heterogeneous data sources. This paper proposes a Dynamic Incentive Sharing Mechanism (DISM) to encourage collaborative intelligence sharing among LLMs in blockchain-based federated learning systems. By providing tamper-proof records of interactions and rewards, blockchain ensures the security and integrity of the learning process. DISM leverages evolutionary game theory to promote cooperation and address challenges such as data poisoning, mitigating disparities in shared intelligence. By calculating each node’s contribution and distributing rewards accordingly, DISM fosters active participation and enhances the overall performance of federated LLM fine-tuning. Experimental simulations and comparative analyses validate the effectiveness of DISM, demonstrating its potential to improve the reliability and efficiency of blockchain-enabled collaborative intelligence sharing.