International Journal of Scientific Research and Engineering Development

By enabling cooperative model training on dispersed datasets without sharing raw data, Federated Learning (FL), a decentralized paradigm, guarantees adherence to privacy laws like the GDPR. Although FL reduces the hazards of centralized data disclosure, it introduces new vulnerabilities through parameter and gradient exchanges. While backdoors, Byzantine attacks, and poisoning are significant security dangers, inference-based approaches such as Deep Leakage from Gradients (DLG) present privacy risks. Secure multi-party computation (SMC), homomorphic encryption (HE), differential privacy, and robust aggregation techniques are some of the countermeasures. Despite progress, there are still unanswered research questions for safe and scalable FL, such as data heterogeneity, privacy vs. model utility trade-offs, and protecting decentralized systems.