A Survey on Secure Aggregation for Privacy-Preserving Federated Learning

Federated learning, an innovative methodology that enables clients to train a global model collectively without disclosing raw data, protects data privacy when it comes to training Machine Learning (ML) models across decentralized devices. This survey provides a concise overview of privacy-preserving federated learning, discussing challenges, techniques, and applications. Various techniques are investigated, including federated learning, differential privacy, homomorphic encryption, Secure Multi-party Computation (SMC), and secret sharing, with a review of their advantages and disadvantages. The survey highlights the importance of secure aggregation methods, emphasizing the necessity for novel algorithms to address challenges such as data heterogeneity and communication latency. Overall, this survey offers valuable insights into privacy-preserving federated learning and its potential impact.