Privacy aware-based federated learning framework for data sharing protection of internet of things devices

Abstract

Federated learning (FL) has emerged as one of the most effective solutions to deal with the rapid utilization of internet of things (IoT) in big data markets. Through FL, local data at each IoT device can be trained locally without sharing the local data to the cloud server. However, this conventional FL may still suffer from privacy leakage when the local data are trained, and the trained model is shared to the cloud server to update the global prediction model. This paper proposes a FL framework with privacy awareness to protect data including the trained model for IoT devices. First, a data/model encryption method using fully homomorphic encryption is introduced, aiming at protecting the data/model privacy. Then, the FL framework for the IoT with the encryption method leveraging logistic regression approach is discussed. Experimental results using random datasets show that the proposed framework can obtain higher global model accuracy (up to 4.84%) and lower global model loss (up to 66.4%) compared with other baseline methods.