Secure Aggregation

Secure aggregation (SA) aims to enable collaborative model training in federated learning (FL) without revealing individual client data to the central server. Current research focuses on developing efficient and robust SA protocols, often incorporating cryptographic techniques like homomorphic encryption or secret sharing, and addressing vulnerabilities to various attacks, including gradient inversion and membership inference. The effectiveness and practicality of SA are crucial for deploying privacy-preserving FL in sensitive applications like healthcare and finance, driving ongoing efforts to improve its efficiency, security, and compatibility with different model architectures and data distributions.