Multi-Level Federated Learning Framework with Group Signatures
Shuangjie Bai, Jinwei Zhu, Xiaoming Hu, Chuang Ma, Ruiling Gao, Qiang Zhou · 2024
Federated learning is a machine learning framework that enables multiple participants to collaboratively train a shared model on local data without having to exchange data directly. This helps maintain data privacy, reduces storage requirements in the data center, and lowers communication costs. In traditional federated learning, participants need to share their local data for model training. This sharing may lead to privacy leakage and security risks. The group signature protocol ensures the anonymity and unlinkability of the participants' identities during the model update process. Additionally, it provides a way to verify whether the model parameters updated by the participants are valid, helping to prevent malicious behavior. In this framework, the user's own information privacy is protected, and a three-layer structure is adopted to further ensure the security and credibility of the model training process.