Research map / Adversarial robustness and security / Certified and empirical robustness
Interpolated Joint Space Adversarial Training for Robust and Generalizable Defenses
Chun Pong Lau, Jiang Liu, Hossein Souri, Wei-An Lin, Soheil Feizi, Rama Chellappa
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 |
|---|---|---|---|
| use exact flow-based manifold instead of GAN for JSA samples (ablation) | AT [JSA-10] · CIFAR-10 (ResNet-18) | standard accuracy: 76.43 -> 82.75 | Improved |
| use exact flow-based manifold instead of GAN for JSA samples (ablation) | AT [JSA-10] · CIFAR-10 (ResNet-18) | average accuracy: 42.07 -> 48.80 | Improved |
| replace PGD samples with JSA samples (ablation) | AT · CIFAR-10 (ResNet-18) | PGD-20 accuracy: 49.85 -> 53.15 | Improved |
| replace PGD samples with JSA samples (ablation) | AT · CIFAR-10 (ResNet-18) | average accuracy: 45.01 -> 48.80 | Improved |
| add plain input mixup after crafting JSA samples (ablation) | AT [JSA-10] · CIFAR-10 (ResNet-18) | average accuracy: 48.80 -> 48.94 | No clear effect |
| add Robust Mixup (perturb after mixup) (ablation) | AT [JSA-10] · CIFAR-10 (ResNet-18) | average accuracy: 48.80 -> 49.25 | Improved |
| increase image-space budget to 16/255 instead of using JSA (ablation) | AT [PGD-10] · CIFAR-10 (ResNet-18) | standard accuracy: 84.15 -> 68.91 | Hurt |
| increase image-space budget to 32/255 (ablation) | AT [PGD-10] · CIFAR-10 (ResNet-18) | average accuracy: 45.01 -> 14.26 | Hurt |
| add JSA to TRADES | TRADES · CIFAR-10 (ResNet-18) | average accuracy: 53.79 -> 54.68 | Improved |
| add JSA to MART | MART · CIFAR-10 (ResNet-18) | average accuracy: 52.36 -> 53.64 | Improved |
| add JSA to MART | MART · CIFAR-10 (ResNet-18) | standard accuracy: 82.81 -> 81.33 | Hurt |
| add JSA to AT | AT · CIFAR-10 (ResNet-18) | Elastic accuracy: 45.16 -> 59.67 | Improved |
Closest papers by meaning
- Dual Manifold Adversarial Robustness: Defense against Lp and non-Lp Adversarial Attacks (NeurIPS 2020 · similarity 0.88)
- Enhancing the Adversarial Robustness via Manifold Projection (AAAI 2025 · similarity 0.87)
- Explaining the role of Intrinsic Dimensionality in Adversarial Training (ICML 2025 · similarity 0.87)
- Exploring Adversarial Robustness of Deep State Space Models (NeurIPS 2024 · similarity 0.86)
- Strong Transferable Adversarial Attacks via Ensembled Asymptotically Normal Distribution Learning (CVPR 2024 · similarity 0.86)
- Adversarial Purification with the Manifold Hypothesis (AAAI 2024 · similarity 0.85)
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.
