A Privacy-Preserving Deep Learning Scheme for Edge-enhanced Smart Homes
Yueyu Dong, Fei Dai, Mingming Qin · 2022 IEEE Intl Conf on Dependable, Autonomic and Secure Computing, Intl Conf on Pervasive Intelligence and Computing, Intl Conf on Cloud and Big Data Computing, Intl Conf on Cyber Science and Technology Congress (DASC/PiCom/CBDCom/CyberSciTech) · 2022
Smart cities make life easier for people. The smart home is an important part of smart cities, and is a typical human-cyber-physical system (HCPS). The application of AI techniques in smart homes requires cloud servers for deep model training, involving the risk of privacy leakage while data transmission to the cloud servers. The data privacy leakage in cyberspace potentially threatens the physical world. The smart home intelligent network gateway is suitable and competent to serve as an edge server. We propose a novel deep model training scheme. Introducing edge computing enables local deep models training, and potentially sensitive local data does not have to leave a trusted environment to avoid the risk of privacy leaks. With the collaboration of edge servers and cloud servers, deep model training is conducted in a federated learning manner. The implementation of privacy-preserving deep model training ensures the application of AI technology in the smart home environment. We have verified the feasibility of the scheme by experiments.