Security, Privacy, and Communication Efficiency in Federated Learning: Issues and Perspectives

Techno-enrichment and its contribution to every sector of human life is raising the expectation of personalized services. These personalized services demand the accumulation of individual’s data for modeling and analytics. This makes users vulnerable to attribute inference attacks. To address these privacy threats, various solutions have been explored, with Federated Learning (FL) emerging as a superior approach. FL enables multiple local devices/organizations to collaboratively train the shared modesl without exposing sensitive information. During this training, all the data remains on local device. As no computer system is safe in all regards as it involves interaction with world, reconstruction of private data is possible. Despite its effectiveness still FL model arises with security and privacy threats. This article prepares a comprehensive study to (i) identify threats to FL from privacy and security point of view, (ii) discuss security and privacy approaches along with communication efficiency, and iii) provide perspective and future research.