Code for paper "Characterizing Adversarial Subspaces Using Local Intrinsic Dimensionality". https://arxiv.org/abs/1801.02613
python train_model.py -d mnist -e 50 -b 128
python craft_adv_samples.py -d cifar -a cw-l2 -b 100
python extract_characteristics.py -d cifar -a cw-l2 -r lid -k 20 -b 100
python detect_adv_examples.py -d cifar -a fgsm -t cw-l2 -r lid
numpy, scipy, tqdm, sklearn, matplotlib, tensorflow >= 1.0, Keras >= 2.0, cleverhans >= 1.0.0 (may need extra change to pass in keras learning rate)