Research map / Adversarial robustness and security / Certified and empirical robustness
Natural Adversarial Examples
Check 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 |
|---|---|---|---|
| Increase ResNet depth from 50 to 152 layers | ResNet · ImageNet-A | accuracy: 2.17% -> 6.05 | Helped |
| Increase ResNet depth from 50 to 152 layers | ResNet · ImageNet-O | AUPR: 16.20 -> 18.00 | Helped |
| Replace ResNet residual blocks with Res2Net v1b blocks | ResNet-50 · ImageNet-A | accuracy: 2.17% -> 14.59% | Helped |
| Replace ResNet residual blocks with Res2Net v1b blocks | ResNet-50 · ImageNet-O | AUPR: 16.20% -> 19.5% | Helped |
| Add Squeeze-and-Excitation | ResNet-50 · ImageNet-A | accuracy: 2.17 -> 6.17 | Helped |
| Adversarially train the model | ResNet-50 · ImageNet-A | accuracy: 2.17% -> 1.68% | Hurt |
| Train on style-transferred ImageNet images | ResNet-50 · ImageNet-A | accuracy: +0.13% | Helped |
| Use AugMix augmentations | ResNet-50 · ImageNet-A | accuracy: 3.8% | Helped |
| Use Cutout augmentation | ResNet-50 · ImageNet-A | accuracy: 4.4% | Helped |
| Exchange feature-map moments between images | ResNet-50 · ImageNet-A | accuracy: 5.5% | Helped |
| Train on convex combinations of images and labels | ResNet-50 · ImageNet-A | accuracy: 6.6% | Helped |
| Use CutMix augmentation | ResNet-50 · ImageNet-A | accuracy: 7.3% | Helped |
Closest papers by meaning
- On Adversarial Training without Perturbing all Examples (ICLR 2024 · similarity 0.89)
- Adversarial Examples Improve Image Recognition (CVPR 2020 · similarity 0.89)
- One Man's Trash Is Another Man's Treasure: Resisting Adversarial Examples by Adversarial Examples (CVPR 2020 · similarity 0.89)
- Detecting Overfitting via Adversarial Examples (NeurIPS 2019 · similarity 0.88)
- Detecting Adversarial Examples Through Image Transformation (AAAI 2018 · similarity 0.88)
- Adversarial Training Can Provably Improve Robustness: Theoretical Analysis of Feature Learning Process Under Structured Data (ICLR 2025 · 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.