Research map / Adversarial robustness and security / Certified and empirical robustness
The Enemy of My Enemy Is My Friend: Exploring Inverse Adversaries for Improving Adversarial Training
Read the paperCheck your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's title only (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 |
|---|---|---|---|
| Use inverse-adversarial training instead of HAT | ResNet-18 · CIFAR-100 | AA robust accuracy: 25.73 vs 23.34 | Helped |
| Generate inverse adversaries in one epoch only | UIAT · CIFAR-10 | AA robust accuracy: 49.11 vs 49.05 | No clear effect |
| Add class-specific universal inverse adversary generation (UAG) (ablation) | IAT · CIFAR-10 | AA robust accuracy: 48.33 -> 47.63 | Hurt |
| Add feature-level regularization (FR) (ablation) | UIAT without IAM · CIFAR-10 | AA robust accuracy: 47.63 -> 48.47 | Helped |
| Add inverse adversary momentum (IAM) (ablation) | UIAT without FR · CIFAR-10 | AA robust accuracy: 47.63 -> 48.83 | Helped |
| Add feature-level regularization and inverse adversary momentum (ablation) | UIAT with UAG · CIFAR-10 | AA robust accuracy: 47.63 -> 49.11 | Helped |
| Add one-off UIAT | N-FGSM · CIFAR-10, epsilon=8/255 | AA robust accuracy: 48.24 -> 49.84 | Helped |
| Add one-off UIAT | RS-FGSM · CIFAR-10, epsilon=8/255 | AA robust accuracy: 46.15 -> 46.31 | Helped |
| Add one-off UIAT | GradAlign · CIFAR-10, epsilon=8/255 | AA robust accuracy: 46.57 -> 46.74 | Helped |
| Increase the UIAT trade-off parameter lambda | UIAT · CIFAR-10 | robust accuracy | Helped |
| Use UIAT rather than TRADES under stronger attacks | ResNet-18 · CIFAR-10, epsilon > 8/255 | AA robust accuracy | Hurt |
Closest papers by meaning
- Improving Adversarial Robustness Requires Revisiting Misclassified Examples (ICLR 2020 · similarity 0.88)
- One Man's Trash Is Another Man's Treasure: Resisting Adversarial Examples by Adversarial Examples (CVPR 2020 · similarity 0.88)
- On Adversarial Training without Perturbing all Examples (ICLR 2024 · similarity 0.88)
- Error Correcting Output Codes Improve Probability Estimation and Adversarial Robustness of Deep Neural Networks (NeurIPS 2019 · similarity 0.88)
- Adversarial Training Can Provably Improve Robustness: Theoretical Analysis of Feature Learning Process Under Structured Data (ICLR 2025 · similarity 0.88)
- A Closer Look at Curriculum Adversarial Training: From an Online Perspective (AAAI 2024 · similarity 0.87)
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.