Research map / Adversarial robustness and security / Black-box and transfer attacks
Enhancing Adversarial Example Transferability With an Intermediate Level Attack
Qian Huang, Isay Katsman, Horace He, Zeqi Gu, Serge Belongie, Ser-Nam Lim
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 ILAP refinement targeted at an intermediate layer | I-FGSM · CIFAR-10 transfer to GoogLeNet (ResNet18 source) | Accuracy after attack: 58.6% -> 35.8% | Improved |
| add ILAP refinement | MI-FGSM · CIFAR-10 transfer to SENet18 (ResNet18 source) | Accuracy after attack: 33.8% -> 30.6% | Improved |
| add ILAP refinement | C&W · CIFAR-10 transfer to SENet18 (ResNet18 source) | Accuracy after attack: 85.4% -> 41.7% | Improved |
| add ILAP refinement | DeepFool · CIFAR-10 transfer to DenseNet121 (ResNet18 source) | Accuracy after attack: 89.1% -> 43.8% | Improved |
| add ILAP refinement | FGSM · CIFAR-10 transfer to SENet18 (ResNet18 source) | Accuracy after attack: 63.6% -> 42.6% | Improved |
| add ILAP refinement | TAP · ImageNet transfer to Inc-v4 | Accuracy after attack: 21.5% -> 16.3% | Improved |
| add ILAP refinement | DI2-FGSM · ImageNet transfer to IncRes-v2 | Accuracy after attack: 54.6% -> 29.3% | Improved |
| replace projection loss ILAP with flexible loss ILAF (tuned alpha) | ILA · CIFAR-10 transfer to SENet18 (ResNet18 source) | Accuracy after attack: 27.5% -> 26.3% | Improved |
| 10 itr ILAP on MI-FGSM output, same model as source (white-box) | MI-FGSM · CIFAR-10 white-box ResNet18 | Accuracy after attack: 5.7% -> 11.3% | Hurt |
| start ILAP from multi-fool perturbation instead of I-FGSM perturbation (ablation) | ILAP · CIFAR-10 (ResNet18 source) | Accuracy after attack | Improved |
| add nonlinearity (activations) after extra linear layers before last layer (ablation) | ILAP on ResNet18 variants · CIFAR-10 (ResNet18 variants) | Accuracy after attack | Improved |
| target later layers with extremely linear decision boundary (ablation) | ILAP · CIFAR-10 (ResNet18 source) | Accuracy after attack | Hurt |
Closest papers by meaning
- ILA-DA: Improving Transferability of Intermediate Level Attack with Data Augmentation (ICLR 2023 · similarity 0.93)
- Yet Another Intermediate-Level Attack (ECCV 2020 · similarity 0.93)
- Transferable Adversarial Perturbations (ECCV 2018 · similarity 0.93)
- Enhancing Adversarial Transferability with Checkpoints of a Single Model's Training (CVPR 2025 · similarity 0.89)
- Boosting Adversarial Transferability through Augmentation in Hypothesis Space (CVPR 2025 · similarity 0.89)
- Enhancing Adversarial Transferability with Adversarial Weight Tuning (AAAI 2025 · similarity 0.89)
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.
