Research map / Adversarial robustness and security / Certified and empirical robustness
Strength-Adaptive Adversarial Training
Chaojian Yu, Dawei Zhou, Li Shen, Jun Yu, Bo Han, Mingming Gong, Nannan Wang, Tongliang Liu
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| add SAAT_up (eps_max=14, rho=1.5) adversary to AT | AT · CIFAR10 PreAct ResNet-18 (last checkpoint) | AutoAttack accuracy: 42.88 -> 46.46 | Improved |
| add SAAT_up (eps_max=14, rho=1.5) adversary to AT | AT · CIFAR10 PreAct ResNet-18 (best checkpoint) | AutoAttack accuracy: 48.23 -> 48.61 | Improved |
| add SAAT_up (eps_max=14, rho=1.5) adversary to AT | AT · CIFAR10 PreAct ResNet-18 (best checkpoint) | Natural accuracy: 82.02 -> 76.80 | Hurt |
| add SAAT_up (eps_max=14, rho=1.7) adversary to AT, tuned rho | AT · CIFAR10 PreAct ResNet-18 (last checkpoint) | AutoAttack accuracy: 42.88 -> 47.17 | Improved |
| add SAAT_down (weaken easy-to-attack samples) to AT (ablation) | AT · CIFAR10 PreAct ResNet-18 (best checkpoint) | AutoAttack accuracy: 48.23 -> 45.54 | Hurt |
| add SAAT_down to AT (ablation) | AT · CIFAR10 PreAct ResNet-18 (best checkpoint) | Natural accuracy: 82.02 -> 82.97 | Improved |
| add SAAT_up to AT on the last checkpoint, narrowing best-last gap (ablation) | AT · CIFAR10 PreAct ResNet-18 | AutoAttack accuracy gap best-last: 48.23 -> 42.88 (AT) vs 48.61 -> 46.46 (SAAT_up eps_max=14) | Improved |
| increase eps_max from 9 to 14 in SAAT_up (rho=1.5) (ablation) | SAAT_up · CIFAR10 PreAct ResNet-18 (best checkpoint) | AutoAttack accuracy: 47.72 -> 48.61 | Improved |
| increase eps_max from 14 to 20 in SAAT_up (rho=1.5) (ablation) | SAAT_up · CIFAR10 PreAct ResNet-18 (best checkpoint) | AutoAttack accuracy: 48.61 -> 45.96 | Hurt |
| increase eps_max from 9 to 18 in SAAT_up (rho=1.5) (ablation) | SAAT_up · CIFAR10 PreAct ResNet-18 (best checkpoint) | PGD-20 accuracy: 53.00 -> 56.49 | Improved |
| increase rho from 1.3 to 1.7 in SAAT_up (eps_max=14) (ablation) | SAAT_up · CIFAR10 PreAct ResNet-18 (best checkpoint) | AutoAttack accuracy: 48.08 -> 48.86 | Improved |
| increase rho from 1.7 to 2.0 in SAAT_up (eps_max=14) (ablation) | SAAT_up · CIFAR10 PreAct ResNet-18 (best checkpoint) | AutoAttack accuracy: 48.86 -> 48.36 | Hurt |
Closest papers by meaning
- One Size Does NOT Fit All: Data-Adaptive Adversarial Training (ECCV 2022 · similarity 0.91)
- l∞-Robustness and Beyond: Unleashing Efficient Adversarial Training (ECCV 2022 · similarity 0.90)
- Boosting Adversarial Robustness with CLAT: Criticality Leveraged Adversarial Training (ICML 2025 · similarity 0.90)
- Adversarial Distributional Training for Robust Deep Learning (NeurIPS 2020 · similarity 0.90)
- Adversarial Training Can Provably Improve Robustness: Theoretical Analysis of Feature Learning Process Under Structured Data (ICLR 2025 · similarity 0.89)
- Taxonomy Driven Fast Adversarial Training (AAAI 2024 · similarity 0.88)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
