Repository Overview

File	Description
main.py	Main train and test file
macer.py	MACER algorithm
model.py	Network architectures
rs/*.py	Randomized smoothing

Make sure you meet package requirements by running:

pip install -r requirements.txt

Example

Here we will show how to train provably l2-robust CIFAR10 and Imgeanet model. We will use σ=0.25 as an example.

Train CIFAR10

python main.py --dataset cifar10 --lr 0.01 --batch_size 64 --training_method macer --sigma 0.25 --lam 12 --gauss_num 16 --label_smoothing True

Train Imagenet

python main.py --dataset imagenet --lr 0.1 --batch_size 256 --data_dir /blob_data/data/imagenet --training_method macer --epochs 120 --sigma 0.25 --lam 6 --gauss_num 2 --label_smoothing True