Research map / Adversarial robustness and security / Black-box and transfer attacks
CDTA: A Cross-Domain Transfer-Based Attack with Contrastive Learning
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 |
|---|---|---|---|
| use CDTA instead of CDA | cross-domain transfer-based attacks · Comic Books | ASR (%): 74.4 -> 83.8 | Helped |
| use CDTA instead of CDA | cross-domain transfer-based attacks · Oxford 102 Flower | ASR (%): 21.3 -> 27.8 | Helped |
| use CDTA instead of CDA | cross-domain transfer-based attacks · BIRDS-400 | ASR (%): 57.4 -> 81.6 | Helped |
| use CDTA instead of CDA | cross-domain transfer-based attacks · Food-101 | ASR (%): 88.5 -> 94.2 | Helped |
| use contrastive spectral training instead of supervised training for the surrogate (ablation) | CDTA · Comic Books | ASR (%): 70.2±2.3; 72.2±2.2; 58.5±2.8; 65.6±1.9 -> 86.0±0.7; 87.0±2.0; 75.4±1.7; 86.8±1.8 | Mixed |
| use contrastive spectral training instead of supervised training for the surrogate (ablation) | CDTA · Oxford 102 Flower | ASR (%): 9.8±0.6; 8.5±1.2; 13.6±1.4; 13.2±1.0 -> 15.8±1.3; 24.8±1.8; 30.6±1.8; 40.0±1.7 | Mixed |
| use contrastive spectral training instead of supervised training for the surrogate (ablation) | CDTA · BIRDS-400 | ASR (%): 71.1±1.8; 69.3±2.1; 47.4±2.0; 71.2±1.3 -> 91.2±1.6; 92.3±0.8; 54.9±3.1; 88.2±1.0 | Mixed |
| use contrastive spectral training instead of supervised training for the surrogate (ablation) | CDTA · Food-101 | ASR (%): 80.8±1.3; 77.9±0.9; 79.5±2.5; 79.6±1.8 -> 94.5±1.1; 94.1±1.1; 93.8±1.3; 94.6±0.6 | Helped |
| use contrastive spectral training instead of conventional SimSiam training for the surrogate (ablation) | CDTA · Comic Books | ASR (%): 25.3±2.4; 28.0±1.9; 25.7±2.6; 25.3±2.1 -> 86.0±0.7; 87.0±2.0; 75.4±1.7; 86.8±1.8 | Helped |
| use contrastive spectral training instead of conventional SimSiam training for the surrogate (ablation) | CDTA · Oxford 102 Flower | ASR (%): 6.8±1.2; 6.1±0.7; 7.7±0.4; 6.2±1.0 -> 15.8±1.3; 24.8±1.8; 30.6±1.8; 40.0±1.7 | Helped |
| use contrastive spectral training instead of conventional SimSiam training for the surrogate (ablation) | CDTA · BIRDS-400 | ASR (%): 19.3±2.4; 21.0±1.5; 7.6±1.0; 23.1±1.9 -> 91.2±1.6; 92.3±0.8; 54.9±3.1; 88.2±1.0 | Helped |
| use contrastive spectral training instead of conventional SimSiam training for the surrogate (ablation) | CDTA · Food-101 | ASR (%): 39.4±3.2; 41.2±1.6; 46.4±2.9; 36.5±1.3 -> 94.5±1.1; 94.1±1.1; 93.8±1.3; 94.6±0.6 | Helped |
Closest papers by meaning
- FACL-Attack: Frequency-Aware Contrastive Learning for Transferable Adversarial Attacks (AAAI 2024 · similarity 0.87)
- PubDef: Defending Against Transfer Attacks From Public Models (ICLR 2024 · similarity 0.85)
- AnchorRep: Defending LLMs Against Cross-Model Adversarial Transfer via Representation Repulsion (NeurIPS 2026 · similarity 0.85)
- Chain of Attack: On the Robustness of Vision-Language Models Against Transfer-Based Adversarial Attacks (CVPR 2025 · similarity 0.85)
- Breaking Barriers in Physical-World Adversarial Examples: Improving Robustness and Transferability via Robust Feature (AAAI 2025 · similarity 0.85)
- Towards Transferable Adversarial Attacks with Centralized Perturbation (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.