PRoVeFL Enhances Private, Robust, and Verifiable Federated Learning

Federated Learning (FL) allows multiple clients to collaboratively train machine learning models while keeping their data localized, thereby boosting user privacy. However, conventional FL setups often depend on a central aggregation server and assume clients are "honest-but-curious," making them vulnerable to server-side inference and client-side poisoning attacks. Existing secure and Byzantine-resilient FL protocols typically face a trade-off between privacy, integrity, and verifiability, often incurring substantial computational and communication overhead due to heavy cryptographic use. To overcome these challenges, researchers have developed PRoVeFL, a modular FL framework designed for Privacy-preserving, Byzantine-Robust, and Verifiable aggregation. PRoVeFL employs multiple servers and leverages multi-key fully homomorphic encryption. Clients encrypt their local model updates and distribute shares to all servers. This enables a hybrid computation model where ciphertext operations are strategically offloaded to the plaintext domain under strict privacy controls, allowing for efficient evaluation of complex statistical aggregation rules. PRoVeFL is compatible with various state-of-the-art Byzantine-robust algorithms and offers significant runtime improvements over previous works while maintaining comparable security.