A Verifiable Federated Learning Algorithm Supporting Distributed Pseudonym Tracking

In the current landscape of federated learning, ensuring the verifiability of aggregated results and guaranteeing the security and authenticity of gradients are essential requirements. Additionally, distributed pseudonymous tracking of participants is required. In this paper, we propose a novel secure data aggregation scheme that simultaneously fulfills all of these requirements. Our scheme incorporates blind signatures and random matrix encoding techniques to ensure the accomplishment of distributed pseudonym tracking and the verifiability of secure gradient aggregation results. Building upon this foundation, the scheme incorporates a commitment system to verify the authenticity of aggregated data returned by the server, thus preventing collusion attacks between participants and the aggregation server. A thorough analysis of accuracy and security confirms that the proposed solution meets the intended design requirements.