11/20/2020: We are developing a new framework for backdoors with FL: Backdoors101. It extends to many new attacks (clean-label, physical backdoors, etc) and has improved user experience. Check it out!

backdoor_federated_learning

This code includes experiments for paper "How to Backdoor Federated Learning" (https://arxiv.org/abs/1807.00459)

All experiments are done using Python 3.7 and PyTorch 1.0.

1. Create a virtual environment in conda conda env create --file environment.yml

2. Activate the conda environment conda activate pytorch_backdoor_FL_test

3. Create a folder for models to be saved mkdir saved_models

4. Open the visdom server for visualizing the training accuracy python -m visdom.server (The server executes in background)

5. Open another terminal, activate the conda environment, and use the following command to train the model python -u training.py 2>&1 | tee -a log_20230310.log

I encourage to contact me ([email protected]) or raise Issues in GitHub, so I can provide more details and fix bugs.

Most of the experiments resulted by tweaking parameters in utils/params.yaml (for images) and utils/words.yaml (for text), you can play with them yourself.

Reddit dataset

• Corpus parsed dataset: https://drive.google.com/file/d/1qTfiZP4g2ZPS5zlxU51G-GDCGGr23nvt/view?usp=sharing
• Whole dataset: https://drive.google.com/file/d/1yAmEbx7ZCeL45hYj5iEOvNv7k9UoX3vp/view?usp=sharing
• Dictionary: https://drive.google.com/file/d/1gnS5CO5fGXKAGfHSzV3h-2TsjZXQXe39/view?usp=sharing