Federated Learning Privacy Protection Scheme Based on Homomorphic Encryption
Yichang Luo, Juan Wang, Yimin Zhou · 2024
Deep learning has been widely applied in various fields such as intelligent driving, military, and medical services. However, with the increasing number of users, the computational and storage requirements of deep learning models have become a challenge for general endpoint devices, making it difficult for general terminal devices to load and apply in the IoT field with low computing power requirements. Federated learning addresses this issue by collectively training models on multiple devices, effectively alleviating the computational limitations of individual user devices. However, federated learning poses privacy leakage risks during the training process, which could expose user privacy data. To address this problem, a federated learning privacy protection scheme based on homomorphic encryption is proposed. This scheme not only resists collusion attacks between the server and learning participants, but also eliminates the need for a trusted third party to generate keys. Therefore, this article's solution has good application prospects in the field of the Internet of Things.