Homomorphic Encryption for Secure Federated Learning: A Narrative Survey

Abstract

Federated learning enables collaborative model training without data centralization but remains vulnerable to information leakage, model manipulation, and aggregation attacks. This narrative review analyses security challenges through the confidentiality, integrity, and availability (CIA) triad, covering architectures, threat models, and privacy-preserving techniques. Differential privacy and secure multi-party computation are briefly noted, while homomorphic encryption is emphasized as the primary cryptographic solution. Security guarantees and performance trade-offs are assessed, identifying homomorphic encryption as a practical approach for strong confidentiality with minimal accuracy loss. The review provides a design-oriented synthesis of current research and forms the conceptual basis for a complementary PRISMA-based systematic review presented in the second part of this study.