In order to see the parameters for the experiments just use main.py -h We use python 3, and we build upon pytorch.

For backdooring (-b option), you can either use "No" backdooring, "Pattern" backdooring for changing the top-left 5*5 to the max intensity as described in the paper, or an index for the specific index of the image from the dataset to behave as a backdoor sample.

• Moran Baruch
• Gilad Baruch
• Yoav Goldberg

If you use this codebase, please cite it as follows:

@article{baruch2019little,
  title={A little is enough: Circumventing defenses for distributed learning},
  author={Baruch, Moran and Baruch, Gilad and Goldberg, Yoav},
  journal={Advances in Neural Information Processing Systems},
  volume={32},
  year={2019}
}