The code for "Improved Deep Leakage from Gradients" (iDLG).

It is widely believed that sharing gradients will not leak private training data in distributed learning systems such as Collaborative Learning and Federated Learning, etc. Recently, Zhu et al. [1] presented an approach which shows the possibility to obtain private training data from the publicly shared gradients. In their Deep Leakage from Gradient (DLG) method, they synthesize the dummy data and corresponding labels with the supervision of shared gradients. However, DLG has difficulty in convergence and discovering the ground-truth labels consistently. In this paper, we find that sharing gradients definitely leaks the ground-truth labels. We propose a simple but reliable approach to extract accurate data from the gradients. Particularly, our approach can certainly extract the ground-truth labels as opposed to DLG, hence we name it Improved DLG (iDLG). Our approach is valid for any differentiable model trained with cross-entropy loss over one-hot labels. We mathematically illustrate how our method can extract ground-truth labels from the gradients and empirically demonstrate the advantages over DLG.

Table 1: Accuracy of the extracted labels for DLG [1] and iDLG. Note that iDLG always extracts the correct label as opposed to DLG which extracts wrong labels frequently.

@article{zhao2020idlg,
  title={iDLG: Improved Deep Leakage from Gradients},
  author={Zhao, Bo and Mopuri, Konda Reddy and Bilen, Hakan},
  journal={arXiv preprint arXiv:2001.02610},
  year={2020}
}

We further leveraged gradient matching to condense the large training set into small synthetic set for efficient deep learning - Dataset Condensation with Gradient Matching [PDF]. [Code]. ICLR 2021 Oral.
Our experiments show that we can condense large training sets into tiny synthetic ones and obtain good generalization ability when train arbitrary randomly initialized deep networks with them.
It's a promising solution to privacy protection and safe federated learning using the synthetic training set.

Table 2: Testing accuracies (%) of ConvNets trained from scratch on 1, 10 or 50 synthetic image(s)/class.