PRoVeFL: Privacy Protection, Byzantine Robustness, and Verifiable Aggregation in Federated Learning

PRoVeFL is a federated learning framework that combines privacy protection, Byzantine robustness, and verifiable aggregation. It uses multi-key fully homomorphic encryption to let clients encrypt and distribute shares of model updates, moves part of the ciphertext computation into the plaintext domain under privacy constraints, supports aggregation algorithms such as Krum and Trimmed Mean, and requires only at least one honest server. Evaluation shows its runtime is up to 100 times and 10 times faster than Prio and ELSA, respectively.