Research map / Adversarial robustness and security / Black-box and transfer attacks
Yet Another Intermediate-Level Attack
Qizhang Li, Yiwen Guo, 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 |
|---|---|---|---|
| use ILA++ (all intermediate discrepancies) instead of ILA | I-FGSM · ImageNet (eps=0.03) | average black-box success rate: 56.45 -> 60.42 | Improved |
| use ILA++ instead of ILA | I-FGSM · CIFAR-100 (eps=0.03) | average black-box success rate: 77.49 -> 79.38 | Improved |
| use ILA++ instead of ILA | I-FGSM · ImageNet (eps=0.1) | average black-box success rate: 94.89 -> 95.08 | No clear effect |
| use ILA++ instead of ILA | TAP · ImageNet (eps=0.03) | average black-box success rate: 62.93 -> 64.21 | Improved |
| use ILA++ instead of ILA | PGD · ImageNet (eps=0.03) | average black-box success rate: 57.64 -> 60.79 | Improved |
| use ILA++ instead of ILA | MI-FGSM · ImageNet (eps=0.03) | average black-box success rate: 54.36 -> 57.54 | Improved |
| add intermediate-level normalization (ablation) | ILA++ · CIFAR-100 (VGG-19 to WRN) | success rate: 85.27 -> 85.53 | Improved |
| set lambda to 10 instead of infinity (ablation) | ILA++ · CIFAR-100 (eps=0.03) | average black-box success rate: 79.38 -> 79.55 | No clear effect |
| use earlier I-FGSM step (p=10) instead of p=100 as guides (ablation) | ILA and ILA++ · CIFAR-100 (VGG-19, WRN) | success rate | Improved |
| ensemble I-FGSM and PGD discrepancies (ablation) | ILA++ · ImageNet (eps=0.03) | average black-box success rate: 60.42 -> 62.82 | Improved |
| use ILA++ instead of ILA | I-FGSM · CIFAR-100 (l2) | average black-box success rate: 73.73 -> 75.78 | Improved |
| use ILA++ instead of ILA | I-FGSM · ImageNet (l2) | average black-box success rate: 74.68 -> 76.73 | Improved |
Closest papers by meaning
- Enhancing Adversarial Example Transferability With an Intermediate Level Attack (ICCV 2019 · similarity 0.93)
- An Intermediate-level Attack Framework on The Basis of Linear Regression (TPAMI 2022 · similarity 0.93)
- Transferable Adversarial Perturbations (ECCV 2018 · similarity 0.91)
- Towards Transferable Black-Box Attack via Minimizing Maximum Model Discrepancy (TPAMI 2026 · similarity 0.91)
- Enhancing Adversarial Transferability with Checkpoints of a Single Model's Training (CVPR 2025 · similarity 0.90)
- 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.
