Maximization of Average Precision for Deep Learning with Adversarial Ranking Robustness

This is the official code for the NeurIPS 2023 spotlight paper "Maximization of Average Precision for Deep Learning with Adversarial Ranking Robustness" by Gang Li, Wei Tong, and Tianbao Yang.

Environment

python==3.8
torch==1.9.1
torchvision==0.10.1
scikit-learn==1.3.0
opencv-python==4.6.0

Training

git clone

git clone https://github.com/GangLii/Adversarial-AP.git

Below is an example of running the method "AdAP_LPN" on CIFAR10 dataset.

CUDA_VISIBLE_DEVICES=0  python ./main_cifar10_resnet18.py \
--method=AdAP_LPN \
--gamma1=0.1 --gamma2=0.9 --Lambda=0.8

Below is an example of running the method "AdAP_LN" on CIFAR100 dataset.

CUDA_VISIBLE_DEVICES=0  python ./main_cifar100_resnet18.py \
--method=AdAP_LN \
--gamma1=0.1 --gamma2=0.9 --Lambda=0.8

List of methods

This is a list of different optimization methods compared in the paper. We summarize them here for reference.

• 'AdAP_LPN' : AdAP_LPN
• 'AdAP_LN' : AdAP_LN
• 'AdAP_LZ' : AdAP_LZ
• 'AdAP_PZ' : AdAP_PZ
• 'AdAP_MM' : AdAP_MM
• 'TRADES' : TRADES
• 'MART' : MART
• 'PGD' : PGD
• 'AP' : AP Max.
• 'CE' : CE Min.

Contact

If you have any questions, please open a new issue or contact Gang Li via [email protected]. If you find this repo helpful, please cite the following paper:

@inproceedings{li2023maximization,
  title={Maximization of Average Precision for Deep Learning with Adversarial Ranking Robustness},
  author={Li, Gang and Tong, Wei and Yang, Tianbao},
  booktitle={Thirty-seventh Conference on Neural Information Processing Systems},
  year={2023}
}