Research map / Adversarial robustness and security / Black-box and transfer attacks
Transferable Adversarial Attacks on Vision Transformers With Token Gradient Regularization
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 |
|---|---|---|---|
| regularize token gradients in ViT intermediate blocks | TGR transfer attacks · ViTs (average; PatchOut) | attack success rate (%): 70.8 -> 76.0 | Helped |
| regularize token gradients in ViT intermediate blocks | TGR transfer attacks · CNNs (average; PatchOut) | attack success rate (%): 42.8 -> 46.7 | Helped |
| regularize token gradients in ViT intermediate blocks | TGR transfer attacks · Adversarially trained CNNs (average; PatchOut) | attack success rate (%): 30.9 -> 33.3 | Helped |
| apply TGR to Attention, QKV, and MLP rather than none (ablation) | TGR on ViT-B/16 · ViTs (average) | attack success rate (%): 56.2 -> 73.6 | Helped |
| add TGR to Attention (ablation) | TGR on ViT-B/16 · ViTs (average) | attack success rate (%): 56.2 -> 67.4 | Helped |
| add TGR to QKV (ablation) | TGR on ViT-B/16 · ViTs (average) | attack success rate (%): 56.2 -> 64.1 | Helped |
| add TGR to MLP (ablation) | TGR on ViT-B/16 · ViTs (average) | attack success rate (%): 56.2 -> 57.3 | Helped |
| include Attention regularization alongside QKV and MLP (ablation) | TGR on ViT-B/16 · ViTs (average) | attack success rate (%): 66.0 -> 73.6 | Helped |
| include QKV regularization alongside Attention and MLP (ablation) | TGR on ViT-B/16 · ViTs (average) | attack success rate (%): 69.3 -> 73.6 | Helped |
| include MLP regularization alongside Attention and QKV (ablation) | TGR on ViT-B/16 · ViTs (average) | attack success rate (%): 69.7 -> 73.6 | Helped |
| increase extreme-token count beyond k=3 (ablation) | TGR on ViT-B/16 · transferability | attack success rate | Hurt |
| increase extreme-token count from 0 to 1 (ablation) | TGR on ViT-B/16 · transferability | attack success rate | Helped |
Closest papers by meaning
- Boosting the Transferability of Adversarial Attack on Vision Transformer with Adaptive Token Tuning (NeurIPS 2024 · similarity 0.93)
- Attacking Transformers with Feature Diversity Adversarial Perturbation (AAAI 2024 · similarity 0.90)
- Harnessing the Computation Redundancy in ViTs to Boost Adversarial Transferability (NeurIPS 2025 · similarity 0.90)
- Improving the Adversarial Transferability of Vision Transformers with Virtual Dense Connection (AAAI 2024 · similarity 0.90)
- Improving Adversarial Transferability on Vision Transformers via Forward Propagation Refinement (CVPR 2025 · similarity 0.88)
- Random Entangled Tokens for Adversarially Robust Vision Transformer (CVPR 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.