Fig. 7From: Iterative and mixed-spaces image gradient inversion attack in federated learningThe ablation experiment for our gradient approximation method on different local epochs. The batch size for CIFAR-10 is 8, while for ImageNet is 4Back to article page