Research map / Adversarial robustness and security / Certified and empirical robustness
Diversifying Counterattacks: Orthogonal Exploration for Robust CLlP Inference
Chengze Jiang, Minjing Dong, Xinli Shi, Jie Gui
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 orthogonal gradient augmentation (OGA) with momentum to TTC counterattack (ablation) | TTC · 16-dataset avg, PGD-10 | robust accuracy (%): 21.43 -> 31.83 | Improved |
| add directional sensitivity score (DSS) to TTC (ablation) | TTC · 16-dataset avg, PGD-10 | clean accuracy (%): 55.66 -> 58.23 | Improved |
| add DSS to TTC (ablation) | TTC · 16-dataset avg, PGD-10 | robust accuracy (%): 21.43 -> 23.37 | Improved |
| add DSS on top of OGA (ablation) | DOC · 16-dataset avg, PGD-10 | clean accuracy (%): 55.38 -> 58.27 | Improved |
| add DSS on top of OGA (ablation) | DOC · 16-dataset avg, PGD-10 | robust accuracy (%): 31.83 -> 31.04 | Hurt |
| replace TTC with DOC | CLIP ViT-B/32 · 16-dataset avg, PGD-10 eps=4/255 | robust accuracy (%): 21.22 -> 31.02 | Improved |
| replace TTC with DOC | CLIP ViT-B/32 · 16-dataset avg, CW eps=4/255 | robust accuracy (%): 20.61 -> 28.18 | Improved |
| replace TTC with DOC | CLIP ViT-B/32 · 16-dataset avg, AutoAttack | robust accuracy (%): 13.0 -> 17.1 | Improved |
| replace TTC with DOC | CLIP ViT-B/32 · EuroSAT, PGD-10 | robust accuracy (%): 14.24 -> 14.49 | No clear effect |
| replace TTC with DOC | CLIP ViT-B/32 · 16-dataset avg, PGD-10 | clean accuracy (%): 55.63 -> 58.26 | Improved |
| counterattack steps N from 1 up to 3 (ablation) | DOC · 12 datasets, PGD-10 and CW | robust accuracy | Improved |
| add DOC as plug-in to adversarially fine-tuned models (TeCoA, PMG, FARE) | adversarial fine-tuning · 16 datasets, PGD-10 eps=1/255 | robust accuracy: 4%-5% over baselines | Improved |
Closest papers by meaning
- R-TPT: Improving Adversarial Robustness of Vision-Language Models through Test-Time Prompt Tuning (CVPR 2025 · similarity 0.89)
- REINFORCE Adversarial Attacks on Large Language Models: An Adaptive, Distributional, and Semantic Objective (ICML 2025 · similarity 0.88)
- TAPT: Test-Time Adversarial Prompt Tuning for Robust Inference in Vision-Language Models (CVPR 2025 · similarity 0.87)
- One Prompt Word is Enough to Boost Adversarial Robustness for Pre-trained Vision-Language Models (CVPR 2024 · similarity 0.87)
- Improving Zero-Shot Adversarial Robustness in Vision-Language Models by Closed-form Alignment of Adversarial Path Simplices (ICML 2025 · similarity 0.87)
- Efficient Robustness Evaluation via Constraint Relaxation (AAAI 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.
