Research map / Adversarial robustness and security / Certified and empirical robustness
Calibrating Uncertainty for Zero-Shot Adversarial CLIP
Wenjing Lu, Zerui Tao, Yuning Qiu, Dongping Zhang, Yang Yang, Qibin Zhao
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 reverse Dirichlet KL alignment to cross-entropy (ablation) | L_ce adversarial fine-tuning · 16 datasets AutoAttack | accuracy: 28.74 -> 30.58 | Improved |
| add reverse Dirichlet KL alignment to cross-entropy (ablation) | L_ce adversarial fine-tuning · 16 datasets clean | accuracy: 43.83 -> 54.17 | Improved |
| add softmax-level KL(p(x)||p(x^a)) to cross-entropy (ablation) | L_ce · 16 datasets AutoAttack | accuracy: 28.74 -> 29.13 | Improved |
| add softmax-level KL(p(x^a)||p(x)) to cross-entropy (ablation) | L_ce · 16 datasets AutoAttack | accuracy: 28.74 -> 28.80 | No clear effect |
| use forward Dirichlet KL instead of reverse (ablation) | UCAT · 16 datasets AutoAttack | accuracy: 30.58 -> 24.36 | Hurt |
| use forward Dirichlet KL instead of reverse (ablation) | UCAT · 16 datasets clean | accuracy: 54.17 -> 36.72 | Hurt |
| apply UCAT to CLIP-B/16 | CLIP-B/16 · 16 datasets AutoAttack | accuracy: 0.01 -> 30.54 | Improved |
| apply UCAT to CLIP-B/16 (clean accuracy cost) | CLIP-B/16 · 16 datasets clean | accuracy: 63.72 -> 52.91 | Hurt |
| apply UCAT to SLIP-B/16 | SLIP-B/16 · 16 datasets AutoAttack | accuracy: 0.02 -> 20.40 | Improved |
| use UCAT instead of TGA-ZSR on COCO multi-label | UCAT · MS-COCO CW eps=1/255 | F1@3: 33.36 -> 37.04 | Improved |
| use UCAT instead of PMG-AFT under AutoAttack | UCAT · 16 datasets AutoAttack | accuracy: 17.88 -> 30.58 | Improved |
| use UCAT on domain-shifted EuroSAT | UCAT vs TGA-ZSR · EuroSAT PGD | accuracy: 11.75 -> 2.99 | 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)
- Improving Zero-Shot Adversarial Robustness in Vision-Language Models by Closed-form Alignment of Adversarial Path Simplices (ICML 2025 · similarity 0.90)
- On the Zero-shot Adversarial Robustness of Vision-Language Models: A Truly Zero-shot and Training-free Approach (CVPR 2025 · similarity 0.89)
- Pre-trained Model Guided Fine-Tuning for Zero-Shot Adversarial Robustness (CVPR 2024 · similarity 0.89)
- Text-Guided Attention is All You Need for Zero-Shot Robustness in Vision-Language Models (NeurIPS 2024 · similarity 0.87)
- CLIPure: Purification in Latent Space via CLIP for Adversarially Robust Zero-Shot Classification (ICLR 2025 · similarity 0.86)
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.
