Lightweight Privacy‐Preserving Learning Schemes*
Yang Yang, Xu Chen, Rui Tan, Yong Xiao · 2021
This chapter proposes three novel privacy-preserving machine learning (ML) schemes on the training or the inference in the context of internet of things (IoT). The first scheme is a lightweight privacy-preserving collaborative learning (PPCL) scheme. The second scheme is also a PPCL approach, in which the fog nodes and the cloud train different stages of a deep neural network, and the data transmitted from an fog node to the cloud is perturbed by Laplacian random noises to achieve 𝜖-differential privacy. The last scheme is designed as a privacy-preserving inference approach. The chapter describes the system model and states the mutual problem. It considers a decentralized ML system consisting of multiple participants and a coordinator. The chapter focuses on the threat and privacy models: honest-but-curious coordinator, and potential collusion between participants and coordinator. It introduces preliminaries about random Gaussian projection.