/robustness

A library for experimenting with, training and evaluating neural networks, with a focus on adversarial robustness.

Primary LanguageJupyter NotebookMIT LicenseMIT

Environment Setting

pip install -e . pip install cox

Train classifier (e.g. Pneumonia classifier)

python train_robust_classifier.py --mimic_path /PATH/TO/MIMIC/DIRECTORY --target labels "Pneumonia" "No Finding" --balance_labels True Right now, only binary classification (ie, two targets) is possible.