Research map / Adversarial robustness and security / Certified and empirical robustness
When Adversarial Training Meets Vision Transformers: Recipes from Training to Architecture
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 ARD and PRM (ablation) | ViT-B adversarial training · CIFAR-10 | AA: 49.06 -> 50.03 | Helped |
| Add ARD and PRM (ablation) | ViT-B adversarial training · Imagenette | AA: 67.00 -> 69.60 | Helped |
| Add ARD and PRM (ablation) | Swin-S adversarial training · CIFAR-10 | AA: 44.88 -> 46.17 | Helped |
| Add ARD and PRM (ablation) | ViT-B adversarial training · ImageNet-1K | AA: 28.81 -> 34.62 | Helped |
| Add ARD and PRM (ablation) | Swin-B adversarial training · ImageNet-1K | AA: 37.91 -> 38.61 | Helped |
| Use SGD + piecewise schedule instead of AdamW + cyclic schedule | ViT-B adversarial training · CIFAR-10 | AA: 46.16 -> 49.06 | Helped |
| Use SGD + piecewise schedule instead of AdamW + piecewise schedule | Swin-B adversarial training · CIFAR-10 | AA: 10.00 -> 46.89 | Helped |
| Add gradient clipping | ViT-B adversarial training · CIFAR-10 | robustness: 10% -> 48.33% | Helped |
| Add gradient clipping | ViT-B adversarial training · Imagenette | robustness: 10% -> 64.20% | Helped |
| Extend training from 40 to 80 epochs (ablation) | ViT-B and Swin-B adversarial training · CIFAR-10 and Imagenette | robustness | Hurt |
| Use CutMix and/or Mixup (ablation) | ViT-B adversarial training · CIFAR-10 | robustness | Helped |
| Use Randaugment (ablation) | ViT-B adversarial training · CIFAR-10 | robustness | Hurt |
Closest papers by meaning
- Vision Transformers Beat WideResNets on Small Scale Datasets Adversarial Robustness (AAAI 2025 · similarity 0.92)
- Harnessing the Computation Redundancy in ViTs to Boost Adversarial Transferability (NeurIPS 2025 · similarity 0.91)
- Random Entangled Tokens for Adversarially Robust Vision Transformer (CVPR 2024 · similarity 0.90)
- Towards Understanding and Improving Adversarial Robustness of Vision Transformers (CVPR 2024 · similarity 0.90)
- Attacking Transformers with Feature Diversity Adversarial Perturbation (AAAI 2024 · similarity 0.90)
- Boosting the Transferability of Adversarial Attack on Vision Transformer with Adaptive Token Tuning (NeurIPS 2024 · similarity 0.89)
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.