Research map / Adversarial robustness and security / Black-box and transfer attacks
A Classifier-Agnostic Zero-Shot Adversarial Attack Detection via CLIP
Hodaya Krakover, Meir Yossef Levi, Eyal Gofer, Guy Gilboa
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 |
|---|---|---|---|
| use PCA aggregation instead of max percentile of prompt similarities (ablation) | A4D · Tiny-ImageNet, DeiT-Small | AU-ROC (avg): 91.53 -> 92.47 | Improved |
| use PCA aggregation instead of min percentile (ablation) | A4D · Tiny-ImageNet, DeiT-Small | AU-ROC (avg): 83.58 -> 92.47 | Improved |
| use PCA aggregation instead of mean percentile (ablation) | A4D · Tiny-ImageNet, DeiT-Small | AU-ROC (avg): 91.04 -> 92.47 | Improved |
| use the CLIP prompt-similarity dictionary with PCA instead of naive distance to trajectory endpoint | A4D · Tiny-ImageNet, DeiT-Small, FGSM | AU-ROC: 93.87 -> 99.24 | Improved |
| use prompt dictionary with PCA instead of naive endpoint distance | A4D · Tiny-ImageNet, DeiT-Small, CW | AU-ROC: 66.31 -> 77.31 | Improved |
| use prompt dictionary with PCA instead of naive endpoint distance | A4D · Tiny-ImageNet, DeiT-Small, DeepFool | AU-ROC: 66.78 -> 78.62 | Improved |
| compare A4D with Wavelet noise baseline on Wide-ResNet average | A4D · Tiny-ImageNet, Wide-ResNet | AU-ROC (avg): 78.36 -> 92.53 | Improved |
| compare A4D with Mahalanobis on ConvNeXt-Tiny average | A4D · Tiny-ImageNet, ConvNeXt-Tiny | AU-ROC (avg): 88.85 -> 84.97 | Hurt |
| compare A4D with attack-agnostic Mahalanobis* on ResNet34 average | A4D · Tiny-ImageNet, ResNet34 | AU-ROC (avg): 38.92 -> 93.32 | Improved |
| detect CW and DeepFool attacks whose perturbations follow image edges | A4D · Tiny-ImageNet, DeiT-Small | AU-ROC: 99.24 -> 77.31 (FGSM vs CW) | Hurt |
| use generic prompt 'A noisy image' instead of curated dictionary (ablation) | A4D · Tiny-ImageNet | AU-ROC | Hurt |
Closest papers by meaning
- CLIP is Strong Enough to Fight Back: Test-time Counterattacks towards Zero-shot Adversarial Robustness of CLIP (CVPR 2025 · similarity 0.90)
- CLIPure: Purification in Latent Space via CLIP for Adversarially Robust Zero-Shot Classification (ICLR 2025 · similarity 0.89)
- On the Zero-shot Adversarial Robustness of Vision-Language Models: A Truly Zero-shot and Training-free Approach (CVPR 2025 · similarity 0.86)
- X-Transfer Attacks: Towards Super Transferable Adversarial Attacks on CLIP (ICML 2025 · similarity 0.85)
- Robust CLIP: Unsupervised Adversarial Fine-Tuning of Vision Embeddings for Robust Large Vision-Language Models (ICML 2024 · similarity 0.85)
- DifAttack: Query-Efficient Black-Box Adversarial Attack via Disentangled Feature Space (AAAI 2024 · similarity 0.84)
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.
