Modeling of Computation Offloading for LEO Satellite-Assisted Federated Learning on Ground-Space Integrated Architecture

Federated learning has gained significant attention as an innovative approach in today's data-driven society. However, traditional federated learning faces challenges such as dependency on a central server and communication delays. Moreover, the feasibility of federated learning in remote areas with limited access to stable ground networks has been largely overlooked. To address these challenges, this paper proposes a novel federated learning architecture that utilizes Low Earth Orbit (LEO) satellites as central server substitutes. LEO satellites offer distributed infrastructure, improved communication capabilities, and enhanced data privacy and security. The proposed architecture aims to overcome the limitations of traditional approaches and enable smooth federated learning in both urban and remote areas. By leveraging the dynamic nature of LEO satellites and introducing offloading techniques, the overall learning delay is optimized. The findings demonstrate the potential of utilizing LEO satellites for federated learning and contribute to the advancement of this field.