A Secure Cloud-Edge Collaborative Logistic Regression Model

Outsourcing logical regression model to the cloud is highly beneficial, yet it will not only cause privacy issues due to the confidentiality of the data, but also cause bandwidth and computing burden for the cloud. In this paper, we design, implement and evaluate a new system that allows secure cloud-edge collaborative logistic regression model (SC-CLRM) based on multi-key fully homomorphic encryption (multi-key FHE). Our system significantly shifts the processing of the cloud to the edge and securely trains logistic regression model among multiple parties. We first propose a new scheme which uses the cloud server and edge node to jointly train logistic regression model over the ciphertext to ensure the security of data and classification model against the cloud server, and data source can go offline after providing the data. We further design security protocols based on multi-key FHE which supports multiple data sources and satisfies the semi-honest model security under outsourcing cloud computing. Performance evaluation demonstrates our new system can securely train the logistic regression model of multiple data sources, as well as share 41.72%~44.92% communication and 53.13%~60.99% computation overhead of the cloud.