Research map / Adversarial robustness and security / Black-box and transfer attacks
Low-Rank and Sparsity Are All You Need: Exploring Robust Hierarchical Latent Subspaces for Transferable Adversarial Attack
Shuangshuang Pu, Wen Yang, Min Li, Guodong Liu, Chris Ding, Di Ming
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 hierarchical mixture of experts (S-M-L) versus no experts (ablation) | LRS-Attack · Inc-v3 surrogate, 6 models | avg ASR: 44.1 -> 49.0 | Improved |
| add hierarchical mixture of experts (S-M-L) versus no experts (ablation) | LRS-Attack · ViT-B surrogate, 6 models | avg ASR: 55.7 -> 73.8 | Improved |
| order experts sparse-middle-deep (S-M-L) instead of low-rank-mixture-sparse (L-M-S) (ablation) | LRS-Attack · Inc-v3 surrogate | avg ASR: 45.4 -> 49.0 | Improved |
| order experts S-M-L instead of L-M-S (ablation) | LRS-Attack · ViT-B surrogate | avg ASR: 72.8 -> 73.8 | Improved |
| use S-M-L instead of all-sparse experts (ablation) | LRS-Attack · Inc-v3 surrogate | avg ASR: 47.8 -> 49.0 | Improved |
| use S-M-L instead of all-low-rank experts (ablation) | LRS-Attack · ViT-B surrogate | avg ASR: 72.9 -> 73.8 | Improved |
| add LRS decomposition versus no decomposition on MI-FGSM | MI-FGSM · Inc-v3 surrogate, 9 CNNs | AvgCNN ASR: 51.7 -> 61.8 | Improved |
| use LRS instead of SVD low-rank attack | MI-FGSM · Inc-v3 surrogate, 9 CNNs | AvgCNN ASR: 55.5 -> 61.8 | Improved |
| use SVD low-rank attack instead of dense baseline | DI-FGSM · Vgg-19 surrogate, 9 CNNs | AvgCNN ASR: 64.6 -> 63.3 | Hurt |
| use LRS instead of SVD under MIM | MIM · DeiT-B surrogate, 8 ViTs | AvgViT ASR: 81.2 -> 86.6 | Improved |
| use LRS instead of SVD under MIM | MIM · ViT-B surrogate, 9 CNNs | AvgCNN ASR: 53.6 -> 57.7 | Improved |
| use LRS instead of dense baseline under ATT | ATT · ViT-B surrogate | AvgDef-CNN ASR: 46.2 -> 45.4 | No clear effect |
Closest papers by meaning
- LRS: Enhancing Adversarial Transferability through Lipschitz Regularized Surrogate (AAAI 2024 · similarity 0.88)
- Transferable Structural Sparse Adversarial Attack Via Exact Group Sparsity Training (CVPR 2024 · similarity 0.88)
- Improving Adversarial Transferability with Local Perturbation Augmentation (CVPR 2026 · similarity 0.88)
- LAFIT: Efficient and Reliable Evaluation of Adversarial Defenses With Latent Features (TPAMI 2023 · similarity 0.87)
- LAFEAT: Piercing Through Adversarial Defenses With Latent Features (CVPR 2021 · similarity 0.87)
- BRUSLEATTACK: A QUERY-EFFICIENT SCORE- BASED BLACK-BOX SPARSE ADVERSARIAL ATTACK (ICLR 2024 · 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.
