Preserving Data security and privacy with federated learning

Federated Learning (FL) has emerged as a promising framework for training machine learning models across distributed clients without sharing raw data. Despite its decentralized architecture, FL remains vulnerable to a range of security threats, including inference, poisoning, backdoor, and Byzantine attacks. This study conducts a systematic review of scientific articles published between ۲۰۱۷ and ۲۰۲۵, analyzing security threats and defense mechanisms in FL. The selected papers are categorized by attack type, defense method, FL algorithm, and application domain. Moreover, using comparative tables and article distribution visualizations, the strengths and limitations of each approach are analyzed. The results reveal that no universal solution exists for securing FL, highlighting the need for multilayer defense strategies, standardized evaluation frameworks, and real-world testing.

کلیدواژه ها:

Federated Learning ، Security ، Privacy Preservation ، Poisoning Attacks ، Secure Aggregation