Credit-based Differential Privacy Stochastic Model Aggregation Algorithm for Robust Federated Learning via Blockchain is submitted in 52nd International Conference on Parallel Processing Proceedings.
There are seven executive documents in our experiment, named CreFlip.py, CreGau.py with two differential privacy mechanisms, and baseline: RSA.py, RSA-flip.py, RSA-Gau.py, SGD.py, SIGNSGD.py.
The options.py has all setting parameters
For executing CreFlip.py: python CreFlip.py --eps=0.4 --lr=0.04 --byzantinue_users=10
In the MNIST dataset around 6 hours, and in the CIFAR dataset around 5 days. (Our code has the potential for further optimization in terms of time, which will be pursued in future work.)
The experiment results(.pkl) are automatically stored in different packages to distinguish different attacks, such as sign_flipping, same_value. We also have two evaluation documents named draw.py and draw_cre.py, representing accuracy and robustness respectively.