Research map / Adversarial robustness and security / Certified and empirical robustness
Robust Learning Meets Generative Models: Can Proxy Distributions Improve Adversarial Robustness?
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 |
|---|---|---|---|
| Add DDPM synthetic data to adversarial training | ResNet-18 · CIFAR-10 ℓ∞ | AutoAttack robust accuracy: 48.7 -> 55.7 | Helped |
| Add DDPM synthetic data to adversarial training | WRN-34-10 · CIFAR-10 ℓ∞ | AutoAttack robust accuracy: 53.1 -> 60.6 | Helped |
| Add DDPM synthetic data to adversarial training | WRN-34-10 · CIFAR-10 ℓ2 | AutoAttack robust accuracy: 73.7 -> 77.8 | Helped |
| Add DDPM synthetic data to adversarial training | PORT · CIFAR-100 ℓ∞ | AutoAttack robust accuracy: 27.8 -> 31.2 | Helped |
| Add DDPM synthetic data to adversarial training | PORT · CIFAR-100 ℓ2 | AutoAttack robust accuracy: 42.1 -> 47.8 | Helped |
| Add DDPM synthetic data to adversarial training | PORT · CelebA ℓ∞ | AutoAttack robust accuracy: 60.2 -> 63.4 | Helped |
| Add DDPM synthetic data to adversarial training | PORT · ImageNet 64×64 ℓ∞ | AutoAttack robust accuracy: 28.1 -> 28.4 | Helped |
| Select synthetic images by synthetic score instead of random sampling (ablation) | PORT · CIFAR-10 ℓ∞ | robust accuracy: 0.5% | Helped |
| Select synthetic images by synthetic score instead of random sampling (ablation) | PORT · CIFAR-10 | robust accuracy: 0.2% | Helped |
| Use diffusion-based synthetic images rather than GAN-generated images | PORT · Multiple datasets | robust accuracy | Helped |
| Use adversarially perturbed robust discrimination rather than ordinary discrimination to assess proxy proximity | ARC · CIFAR-10 generative-model comparison | ranking agreement with robustness transfer | Helped |
| Add synthetic data to randomized smoothing training | PORT · CIFAR-10 certified robust accuracy | certified robust accuracy: 7.6% | Helped |
Closest papers by meaning
- Provable Robust Overfitting Mitigation in Wasserstein Distributionally Robust Optimization (ICLR 2025 · similarity 0.88)
- Diffusion Models are Certifiably Robust Classifiers (NeurIPS 2024 · similarity 0.88)
- Robust Classification via a Single Diffusion Model (ICML 2024 · similarity 0.88)
- On the Scalability of Certified Adversarial Robustness with Generated Data (NeurIPS 2024 · similarity 0.88)
- Adversarial Training Can Provably Improve Robustness: Theoretical Analysis of Feature Learning Process Under Structured Data (ICLR 2025 · similarity 0.88)
- GREAT Score: Global Robustness Evaluation of Adversarial Perturbation using Generative Models (NeurIPS 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.