Research map / Adversarial robustness and security / Black-box and transfer attacks
Inter-Class Topology Alignment for Efficient Black-Box Substitute Attacks
Lingzhuang Meng, Mingwen Shao, Yuanjian Qiao, Wenjie Liu
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 relative position loss on PES to distillation loss (DIST metric for KD) (ablation) | ICTA · Target attack PGD | attack success rate: 57.67 -> 79.34 | Improved |
| use DIST metric instead of MSE in distillation loss (ablation) | ICTA · Target attack PGD | attack success rate: 46.14 -> 57.67 | Improved |
| explore decision space with PES (full gradient, unconstrained) rather than BIM adversarial or clean samples (ablation) | ICTA · CIFAR100 ResNet50 | attack success rate | Improved |
| apply ICTA vs LCA to non-target attack on CIFAR10 VGG16 | ICTA · CIFAR10 VGG16 non-target PGD | attack success rate: 97.01 -> 96.48 | Hurt |
| ICTA on larger datasets (TinyImageNet) rather than CIFAR10 | ICTA · data-free non-target PGD | attack success rate: 70.58 -> 93.39 on TinyImageNet | Improved |
| ICTA when only label outputs of target are available | ICTA · CIFAR100 label-based target attack | attack success rate: 49.85 -> 61.71 | Improved |
| compare substitute accuracy to attack success | ICTA · Full-data CIFAR100 ResNet50 | substitute accuracy | No clear effect |
Closest papers by meaning
- AGS: Affordable and Generalizable Substitute Training for Transferable Adversarial Attack (AAAI 2024 · similarity 0.84)
- Highly Imperceptible Black-Box Graph Injection Attacks with Reinforcement Learning (AAAI 2025 · similarity 0.83)
- Adversarial Attacks against Closed-Source MLLMs via Feature Optimal Alignment (NeurIPS 2025 · similarity 0.83)
- AnchorRep: Defending LLMs Against Cross-Model Adversarial Transfer via Representation Repulsion (NeurIPS 2026 · similarity 0.83)
- Crossfire: An Elastic Defense Framework for Graph Neural Networks Under Bit Flip Attacks (AAAI 2025 · similarity 0.83)
- Open-Set Cross-Network Node Classification via Unknown-Excluded Adversarial Graph Domain Alignment (AAAI 2025 · similarity 0.82)
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.
