A Secure Enhancement Scheme for Interworking Privacy-Preserving Multi-Party Collaborative Modeling

Multi-party collaborative modeling allows different participants to build machine learning models without revealing the local data. Federated Learning (FL) is currently the main technique to achieve multi-party collaborative modeling. However, FL faces a series of security and privacy issues (e.g., gradient leakage attacks). We present a new federated learning framework that incorporates fine-grained access control for privacy-preserving collaborative modeling by incorporating identity-based encryption (IBE), identity-based broadcast encryption (IBBE), and the trusted execution environment (TEE) technique. In this framework, participants are coordinated through a key generation center. Our framework enables secure broadcasting of the global parameters and fine-grained access control of the intermediate parameters during the FL process. Compared with prior FL frameworks, our framework does not require the communication channel to be secure and there is no additional key management overhead. Besides, in our framework, there is no need to establish a connection between the clients and server in advance. We conducted a series of experiments to prove that our solution is practical and efficient.