Federated Learning with Differential Privacy + MLflow & Optuna
CNN Framework with extensive hyperparamter optimization using optuna and model tracking with mlflow. Best performing model is use as baseline for Federated Learning and Federated Learning with Differential Privacy.
Instructions for use:
- create your virtual env -> activate
- pip install -r requirements.txt
- source folder is the brain
- data folder has a sample dataset for less intensive experiments
- helpers: any helper function to prepare the dataset
- src:
- 3 folders centralized, federated, and federated_dp holds the codes corresponding to the 3 approaches
- models - final model parameters
- run predict.py to evaluate on test data