Privacy-Preserving Edge Intelligent Computing Based on Federated Learning

The rapid development of the Internet of things industry makes the application of edge intelligent computing more and more widely, but it also exposes serious problems of data security and privacy disclosure. Federated learning can cooperate with multiple terminals to complete collaborative training without local data, which can reduce the risk of privacy disclosure of edge intelligent computing. However, it is difficult to resist the latest privacy attacks such as model extraction and model inversion, and the traditional privacy protection protocol can not be applied to edge devices with limited computing and communication resources. Therefore, this paper proposes an edge intelligent computing privacy protection method based on federated learning. Based on secret sharing and gradient mask, a lightweight privacy protection protocol is designed to protect the privacy of device data with less computing overhead. A periodic update strategy is proposed to reduce the communication overhead. Compared with similar methods on public data sets, the results show that this method performs better in model accuracy, communication overhead and operation efficiency, and can effectively resist gradient disclosure attack.