Research map / Adversarial robustness and security / Black-box and transfer attacks
Boosting Adversarial Transferability using Dynamic Cues
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 temporal prompts to Deit-T for FGSM | Deit-T image surrogate · TimesFormer / UCF | adversarial accuracy (%): 75.1 -> 64.5 | Helped |
| Add temporal prompts to Deit-B for FGSM | Deit-B image surrogate · TimesFormer / K400 | adversarial accuracy (%): 57.5 -> 48.6 | Helped |
| Add temporal prompts to DINO for MIM | DINO image surrogate · TimesFormer / UCF | adversarial accuracy (%): 71.3 -> 47.7 | Helped |
| Add temporal prompts to DINO for DIM | DINO image surrogate · TimesFormer / UCF | adversarial accuracy (%): 64.9 -> 45.2 | Helped |
| Add multiscale temporal prompts to Deit-B | Deit-B image surrogate · ImageNet val (5k) | fool rate (%): 80.10 -> 86.64 | Helped |
| Add multiscale temporal prompts to CLIP | CLIP image surrogate · ImageNet val (5k) | fool rate (%): 80.10 -> 55.34 | Hurt |
| Increase temporal prompt count from 8 to 32 (ablation) | Deit-B DIM surrogate · TimesFormer / UCF | adversarial accuracy (%): 55.5 -> 21.5 | Helped |
| Increase temporal prompt count from 8 to 32 (ablation) | DINO DIM surrogate · TimesFormer / UCF | adversarial accuracy (%): 45.2 -> 30.2 | Helped |
| Add spatial-resolution variation to temporal prompts (ablation) | DeiT-B surrogate · TimesFormer / UCF | adversarial accuracy (%): 55.5 -> 53.9 | Helped |
| Fully fine-tune the image backbone rather than training prompts only (ablation) | DINO surrogate with temporal prompts · TimesFormer / UCF | adversarial accuracy (%): 45.2 -> 56.3 | Hurt |
| Fully fine-tune the image backbone rather than training prompts only (ablation) | Deit-B surrogate with temporal prompts · TimesFormer / UCF | adversarial accuracy (%): 55.5 -> 59.3 | Hurt |
| Use DINO rather than CLIP as the image surrogate | Image-to-video DIM attack · TimesFormer / UCF | adversarial accuracy (%): 73.4 -> 45.2 | Helped |
Closest papers by meaning
- Boosting the Transferability of Adversarial Attack on Vision Transformer with Adaptive Token Tuning (NeurIPS 2024 · similarity 0.90)
- Learning to Transform Dynamically for Better Adversarial Transferability (CVPR 2024 · similarity 0.89)
- Harnessing the Computation Redundancy in ViTs to Boost Adversarial Transferability (NeurIPS 2025 · similarity 0.89)
- Enhancing Adversarial Transferability with Adversarial Weight Tuning (AAAI 2025 · similarity 0.89)
- Enhancing Adversarial Transferability with Checkpoints of a Single Model's Training (CVPR 2025 · similarity 0.88)
- I2VGuard: Safeguarding Images against Misuse in Diffusion-based Image-to-Video Models (CVPR 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.