RR-LADP: A Privacy-Enhanced Federated Learning Scheme for Internet of Everything

Abstract

While the widespread use of ubiquitously connected devices in Internet of Everything (IoE) offers enormous benefits, it also raises serious privacy concerns. Federated learning, as one of the promising solutions to alleviate such problems, is considered as capable of performing data training without exposing raw data that kept by multiple devices. However, either malicious attackers or untrusted servers can deduce users' privacy from the local updates of each device. Previous studies mainly focus on privacy-preserving approaches inside the servers, which require the framework to be built on trusted servers. In this article, we propose a privacy-enhanced federated learning scheme for IoE. Two mechanisms are adopted in our approach, namely the randomized response (RR) mechanism and the local adaptive differential privacy (LADP) mechanism. RR is adopted to prevent the server from knowing whose updates are collected in each round. LADP enables devices to add noise adaptively to its local updates before submitting them to the server. Experiments demonstrate the feasibility and effectiveness of our approach.