Revisiting Adversarial Robustness Distillation from the Perspective of Robust Fairness

Code for NeurIPS 2023 "Revisiting Adversarial Robustness Distillation from the Perspective of Robust Fairness".

Environment

• Python (3.9.12)
• Pytorch (1.11.0)
• torchvision (0.12.0)
• CUDA
• AutoAttack

Content

• ./models: Models used for distillation.
• Fair-ARD.py: Fair Adversarial Robustness Distillation.
• Fair-IAD.py: Fair Introspective Adversarial Distillation.
• Fair-RSLAD.py: Fair Robust Soft Label Adversarial Distillation.
• Fair-MTARD.py: Fair Multi-Teacher Adversarial Robustness Distillation.
• eval.py: Evaluate the average robustness and worst-class robustness of the model.

Run

• Fair-ARD

CUDA_VISIBLE_DEVICES='0' python Fair-ARD.py --teacher_path INSERT-YOUR-TEACHER-PATH --beta 2.0

• Fair-IAD

CUDA_VISIBLE_DEVICES='0' python Fair-IAD.py --teacher_path INSERT-YOUR-TEACHER-PATH --beta 2.0

• Fair-RSLAD

CUDA_VISIBLE_DEVICES='0' python Fair-RSLAD.py --teacher_path INSERT-YOUR-TEACHER-PATH --beta 2.0

• Fair-MTARD

CUDA_VISIBLE_DEVICES='0' python Fair-MTARD.py --adv_teacher_path INSERT-YOUR-ADV-TEACHER-PATH --nat_teacher_path INSERT-YOUR-NAT-TEACHER-PATH --beta 2.0

• Evaluation

CUDA_VISIBLE_DEVICES='0' python eval.py --model_path INSERT-YOUR-MODEL-PATH

Pre-trained Models

• The teacher models and pre-trained models can be downloaded from the Google Drive

Reference Code

[1] ARD: https://github.com/goldblum/AdversariallyRobustDistillation

[2] IAD: https://github.com/ZFancy/IAD

[3] RSLAD: https://github.com/zibojia/RSLAD

[4] MTARD: https://github.com/zhaoshiji123/MTARD

[5] GAIRAT: https://github.com/zjfheart/Geometry-aware-Instance-reweighted-Adversarial-Training