Scholarlink

The surge in affordable satellite launch options has significantly expanded the number of low-earth orbit (LEO) satellites, enabling an unparalleled level of instantaneous data availability for a wide range of applications. The utility of these applications depends on the continuous and adept training of machine learning (ML) models. However, traditional ground-training methods suffer from limited bandwidth and inconsistent satellite communications. To overcome these hurdles, this paper presents SkyChain, a novel blockchain-empowered federated learning (FL) framework specifically designed for LEO satellites. SkyChain is built with a proof-of-stake (PoS)-based consensus protocol and includes a mechanism for validating local model updates, thus creating a secure and efficient decentralized ML ecosystem in the spatial domain. This approach notably advances real-time communication, reinforces data reliability, and provides transparent and secure model updates. We detail the architecture, design, and consensus mechanism implementation of SkyChain. We demonstrate the stability and efficiency of the proposed PoS mechanism for FL-based LEO satellites through comparative performance analysis.

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

SkyChain: Secure Federated Learning with Model Validation in LEO Satellite Constellations

  • 会议论文

  • 发表日期：

    2026年

查看全文
  • Zenebe Melesew Yetneberk,
  • Tong-Xing Zheng,
  • Xinji Wang,
  • Haiyang Cao,
  • Xin Chen,
  • Lei Lei,
  • Ying Ju

摘要

The surge in affordable satellite launch options has significantly expanded the number of low-earth orbit (LEO) satellites, enabling an unparalleled level of instantaneous data availability for a wide range of applications. The utility of these applications depends on the continuous and adept training of machine learning (ML) models. However, traditional ground-training methods suffer from limited bandwidth and inconsistent satellite communications. To overcome these hurdles, this paper presents SkyChain, a novel blockchain-empowered federated learning (FL) framework specifically designed for LEO satellites. SkyChain is built with a proof-of-stake (PoS)-based consensus protocol and includes a mechanism for validating local model updates, thus creating a secure and efficient decentralized ML ecosystem in the spatial domain. This approach notably advances real-time communication, reinforces data reliability, and provides transparent and secure model updates. We detail the architecture, design, and consensus mechanism implementation of SkyChain. We demonstrate the stability and efficiency of the proposed PoS mechanism for FL-based LEO satellites through comparative performance analysis.