Research map / Adversarial robustness and security / Certified and empirical robustness
You Don't Protect if You Don't Expect: Breaking the Key Assumption behind CLIP's Test-Time Defenses
Ruize Zhang, Yu Li, Zhang Wan, Juan Cao, Jie Zhang, Sheng Tang
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (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 CLIP-MAD on top of full-gradient attack | TTC defense · PGD-10 eps=4/255, 13 datasets | average robust accuracy (lower means stronger attack): 11.0 -> 4.7 | Improved |
| add CLIP-MAD on top of full-gradient attack | DOC defense · PGD-10 eps=4/255, 13 datasets | average robust accuracy (lower means stronger attack): 19.5 -> 9.4 | Improved |
| add CLIP-MAD on top of static attack | R-TPT defense · PGD-10 eps=4/255, 13 datasets | average robust accuracy (lower means stronger attack): 37.5 -> 0.7 | Improved |
| add CLIP-MAD on top of full-gradient attack | TAPT-VLI defense · PGD-10 eps=4/255, 13 datasets | average robust accuracy (lower means stronger attack): 0.3 vs 0.4 | No clear effect |
| use CLIP-MAD instead of static attack | CLIPure-Cos defense · PGD-10 eps=1/255, 13 datasets | average robust accuracy (lower means stronger attack): 6.6 -> 4.8 | Improved |
| use CLIP-MAD instead of static attack | CLIPure-Diff defense · PGD-10 eps=1/255, 13 datasets | average robust accuracy (lower means stronger attack): 7.3 -> 6.0 | Improved |
| use full-gradient attack instead of static attack | CLIPure-Cos defense · PGD-10 eps=1/255, 13 datasets | average robust accuracy (lower means stronger attack): 6.6 -> 78.0 | Hurt |
| use CLIP-MAD instead of static attack | R-TPT defense · APGD restart=2 eps=4/255, ImageNet | average robust accuracy (lower means stronger attack): 37.6 -> 0.6 | Improved |
| use CLIP-MAD instead of full-gradient attack | TTC defense · PGD restart=3 EOT-10 eps=4/255, CIFAR10 | average robust accuracy (lower means stronger attack): 12.0 -> 3.9 | Improved |
Closest papers by meaning
- CLIP is Strong Enough to Fight Back: Test-time Counterattacks towards Zero-shot Adversarial Robustness of CLIP (CVPR 2025 · similarity 0.93)
- CLIPure: Purification in Latent Space via CLIP for Adversarially Robust Zero-Shot Classification (ICLR 2025 · similarity 0.88)
- TAPT: Test-Time Adversarial Prompt Tuning for Robust Inference in Vision-Language Models (CVPR 2025 · similarity 0.87)
- On the Adversarial Risk of Test Time Adaptation: An Investigation into Realistic Test-Time Data Poisoning (ICLR 2025 · similarity 0.86)
- X-Transfer Attacks: Towards Super Transferable Adversarial Attacks on CLIP (ICML 2025 · similarity 0.85)
- Closing the Gap: Achieving Better Accuracy-Robustness Tradeoffs against Query-Based Attacks (AAAI 2024 · similarity 0.85)
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.
