Fig. 2: Accuracy and system performance comparison of Lancelot.
From: Towards compute-efficient Byzantine-robust federated learning with fully homomorphic encryption
a, Accuracy comparison of Lancelot under targeted and untarget attacks across the MNIST, FMNIST, CIFAR-10 and SVHN datasets. b, Execution time per round of Lancelot, including both computation and communication time, among three methods across three datasets. Speed-up refers to the acceleration ratio between Lancelot and OpenFHE. MK, Multi-Krum.
