The environment is available in env/fl_proj_env.yml
. I currently use Conda to manage the environment.
The entrance to our implementation is defense_trial.py
. Using python defense_trial.py
is sufficient to begin and I have tested on Ubuntu 20.08. To customize the arguments, please refer to utils/options.py
for further information.
- The project has so far fulfilled the original design target. To be specific, we aim to defend GRNN attack, and our DP implementation indeed achieve the goal, with reasonable accuracy loss.
- I didn't modify the structure of GRNN, and there should be consequent work on refining it. In future I can look into some more advanced framework.
- The current implementation of DP is not audited.