Privacy-Preserving Deep Learning in Internet of Healthcare Things with Blockchain-Based Incentive

Privacy-preserving deep learning has drawn tremendous attention recently, especially in the IoHT-enabled medical field. As a representative, federated learning can guarantee the privacy of training data and training models, but there are still many security issues that are ignored. During the training process, the content of parameters may be tampered with to affect the overall accuracy, and the parameter server may also be malicious. In this paper, we propose a blockchain architecture to solve these problems, which uses blockchain-based payment incentive method to force miners and medical institutions to behave honestly, thereby speeding up convergence. In addition, considering that the miners are disconnected in the real network environment, which leads to the interruption of the consensus protocol and affects the convergence speed, we design the Robust Proof-of-Stake (RPoS) consensus based on PVSS to solve this problem. Experiments show that the incentive mechanism we design can improve the accuracy of predictions and reduce the possibility of dishonesty among participants.