Federated Learning-Based Privacy Protection for IoT-based Smart Healthcare Systems
Fan Jiang, Zidong Chen, Lei Liu, Junxuan Wang · 2023
In this paper, we propose a Federated Privacy-Enhanced Healthcare (FPEH) learning framework based on Gaussian differential privacy and Paillier homomorphic encryption to enhance data privacy protection for IoT-based smart healthcare systems. To address the data privacy issues of the clients in this scenario, we first apply a Gaussian differential privacy-based mechanism to add noise to the parameters in the client models. Secondly, a constant factor is initially introduced to enhance the randomness and unpredictability of the Gaussian noise. Furthermore, a gradient clipping mechanism is also adopted to ensure the effectiveness of the client model after adding noise. Finally, we employ Paillier homomorphic encryption to encrypt the noisy model parameters, further ensuring the security of the original data. We evaluate the proposed method by applying two datasets, namely MNIST and PneumoniaMNIST. Simulation results demonstrate that the proposed method can protect the privacy of the original data from being leaked and achieve a similar test accuracy level as the comparison algorithms.