Research map / Adversarial robustness and security / Black-box and transfer attacks
An Intermediate-level Attack Framework on The Basis of Linear Regression
Yiwen Guo, Qizhang Li, Wangmeng Zuo, Hao Chen
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 linear-regression intermediate-level guidance (RR, ILA++) on top of I-FGSM | I-FGSM · ImageNet untargeted l_inf eps=8/255 | avg attack success rate: 27.46% -> 63.38% | Improved |
| replace ILA with the regression-based ILA++ | I-FGSM+ILA · ImageNet untargeted l_inf eps=8/255 | avg attack success rate: 60.02% -> 63.38% | Improved |
| maximize intermediate-level feature magnitude instead of normalizing the discrepancy (ablation) | I-FGSM+RR · ImageNet untargeted l_inf eps=16/255 | avg attack success rate: 31.48% -> 88.57% | Improved |
| maximize intermediate-level feature magnitude instead of normalizing the discrepancy (ablation) | I-FGSM+RR · ImageNet untargeted l2 eps=15 | avg attack success rate: 63.54% -> 87.66% | Improved |
| use baseline-derived adversarial directional guide instead of random input-space guide (ablation) | I-FGSM+RR · ImageNet untargeted l_inf eps=8/255 | avg attack success rate: 21.25% -> 63.38% | Improved |
| draw random guide in input space instead of feature space (ablation) | Rand+RR · ImageNet untargeted l_inf eps=8/255 | avg attack success rate: 12.26% -> 21.25% | Improved |
| use SVR instead of ridge regression | I-FGSM+RR · ImageNet untargeted l_inf eps=8/255 | avg attack success rate: 63.38% -> 64.01% | Improved |
| use SVR instead of ridge regression | I-FGSM+RR · ImageNet untargeted l2 eps=10 | avg attack success rate: 74.78% -> 75.31% | Improved |
| use ElasticNet instead of ridge regression | I-FGSM+RR · ImageNet untargeted l_inf eps=8/255 | avg attack success rate: 63.38% -> 50.07% | Hurt |
| decrease ElasticNet lambda1 from 1.0 to 0.05 (ablation) | I-FGSM+ElasticNet · ImageNet untargeted l_inf eps=8/255 | avg attack success rate: 31.97% -> 50.07% | Improved |
| decrease ElasticNet lambda1 from 1.0 to 0.05 (source model ResNet-50) (ablation) | I-FGSM+ElasticNet · ImageNet untargeted l_inf eps=8/255 (source model) | avg attack success rate: 98.16% -> 99.92% | Improved |
| collect guides from 10 random-init PGD runs with SVR | I-FGSM+SVR · ImageNet untargeted l_inf eps=8/255 | avg attack success rate: 64.01% -> 66.14% | Improved |
Closest papers by meaning
- Transferability Bound Theory: Exploring Relationship between Adversarial Transferability and Flatness (NeurIPS 2024 · similarity 0.89)
- Enhancing Adversarial Transferability with Adversarial Weight Tuning (AAAI 2025 · similarity 0.88)
- PubDef: Defending Against Transfer Attacks From Public Models (ICLR 2024 · similarity 0.87)
- Boosting Adversarial Transferability through Augmentation in Hypothesis Space (CVPR 2025 · similarity 0.87)
- Rethinking Model Ensemble in Transfer-based Adversarial Attacks (ICLR 2024 · similarity 0.87)
- LRS: Enhancing Adversarial Transferability through Lipschitz Regularized Surrogate (AAAI 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.
