Verifiable Federated Learning With Privacy-Preserving Data Aggregation for Consumer Electronics

With the development of information technology, massive and heterogeneous consumer electronic products can access the network. These products may engage third-party servers for federated learning and generating more accurate models, where they can monitor, collect and aggregate various data from households almost in real-time. Even though federated learning can update participant parameter data without collecting their raw data, prior research revealed that the shared gradients still retain sensitive information from the training set. Meanwhile, malicious third-party aggregation servers may return forged aggregated gradients, and lightweight execution of the entire solution needs to be ensured during the aggregation process. This paper demonstrates a verifiable federated learning scheme supporting secure data aggregation without using bilinear groups (FLVA) to address these issues. Particularly, to solve the issue of private key leakage in the gradient aggregation process on electronic product data, a three-party key negotiation protocol is developed. The private gradients are uploaded and aggregated in ciphertext format, which ensures the privacy of the electronic product gradient. Security analysis showed that our FLVA system can effectively protect the security and privacy of the private gradients. Finally, the experimental results showed that compared with existing solutions, the proposed scheme is more efficient and practical.