Privacy Preservation using Federated Learning and Homomorphic Encryption: A Study

The resurgence of Artificial Intelligence (AI) and its broader adoption into major fields has revolutionized the operations of organizations and applications. AI depends on Machine Learning (ML) models that are suitably trained using data generated by global users. Preserving the privacy and security of global user’s data in ML and Deep Learning (DL) has emerged as a major concern. The advent of Federated Learning (FL) - an ML paradigm, has emerged as a solution for the privacy concerns related to traditional training models. FL is a decentralized training method that enables computation of well-trained models over remote data sources located at different geographical locations without sharing the training data to a centralized location. The aim of this study is to provide a detailed understanding of FL. This work focuses on the significant challenges and attacks associated with data privacy in FL and also the solutions that are available in addressing these attacks. This paper highlights the features and importance of Homomorphic Encryption (HE) towards privacy preservation in FL. This work also discusses open challenges and potential future scope and applications of FL.