Research map / Adversarial robustness and security / Black-box and transfer attacks
A New Ensemble Adversarial Attack Powered by Long-Term Gradient Memories
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 SMBEA instead of Liu's Ensemble Attack | black-box pixel-to-pixel ensemble attacks · Cityspaces, Facades, Google Satellite, and LSUN'17 | performance drop (MSE for image translation; CC for saliency): Cityspaces: +0.0118 vs +0.0148; +0.0194 vs +0.0213; +0.0230 vs +0.0264. Facades: +0.0129 vs +0.0155; +0.0165 vs +0.0230; | Helped |
| Increase the number of mini-batches | SMBEA · Cityspaces, Facades, Google Satellite, and LSUN'17 | performance drop: Cityspaces: +0.0148; +0.0213; +0.0264. Facades: +0.0155; +0.0230; +0.0275. Google Satellite: +0.0108; +0.0137; +0.0145. | Helped |
| Add feature-space ensemble strategies to output-space ensembling (ablation) | SMBEA intra-batch attack · Pixel-to-pixel tasks | fooling ability | Helped |
| Use the third feature-space ensemble strategy rather than the other strategies (ablation) | SMBEA intra-batch attack · Pixel-to-pixel tasks | fooling ability | Helped |
| Select deeper feature layers (ablation) | SMBEA feature-space ensemble · Pixel-to-pixel tasks | fooling ability | Helped |
| Add intra-batch short-term gradient momentum (ablation) | SMBEA · Black-box transfer setting | transferability | Helped |
| Add inter-batch long-term gradient memories (ablation) | SMBEA · Black-box transfer setting | transferability | Helped |
| Use SMBEA rather than MIM or I-FGSM | attacks against adversarially trained models · LSUN'17, GazeGAN and SAM-ResNet | CC performance drop: GazeGAN, 3 batches: -0.1315 -> -0.2190; 5 batches: -0.1582 -> -0.2731; 7 batches: -0.2063 -> -0.3255. SAM-ResNet, 3 batc | Helped |
Closest papers by meaning
- Rethinking Model Ensemble in Transfer-based Adversarial Attacks (ICLR 2024 · similarity 0.91)
- Ensemble Adversarial Training: Attacks and Defenses (ICLR 2018 · similarity 0.90)
- Ensemble Diversity Facilitates Adversarial Transferability (CVPR 2024 · similarity 0.89)
- Enhancing Adversarial Transferability with Checkpoints of a Single Model's Training (CVPR 2025 · similarity 0.89)
- Understanding Model Ensemble in Transferable Adversarial Attack (ICML 2025 · similarity 0.88)
- Strong Transferable Adversarial Attacks via Ensembled Asymptotically Normal Distribution Learning (CVPR 2024 · similarity 0.87)
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.