Research map / Adversarial robustness and security / Black-box and transfer attacks
Once a MAN: Towards Multi-Target Attack via Learning Multi-Target Adversarial Network Once
Jiangfan Han, Xiaoyi Dong, Ruimao Zhang, Dongdong Chen, Weiming Zhang, Nenghai Yu, Ping Luo, Xiaogang Wang
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 |
|---|---|---|---|
| train with random target labels (multi-target) instead of one fixed target (ablation) | MANr · CIFAR10 Res32 to VGG16 transfer | attack success rate (%): 58.71 -> 67.69 | Improved |
| jointly train against VGG16 and ResNet32 (model ensemble) (ablation) | MANr · CIFAR10 Res32 to VGG19 transfer | attack success rate (%): 72.26 -> 98.07 | Improved |
| jointly train against VGG16 and ResNet32 (model ensemble) (ablation) | MANc · CIFAR10 Res32 to Res14 transfer | attack success rate (%): 81.01 -> 96.79 | Improved |
| use the channel-recalibration label fusion (MANr) instead of concatenation (MANc) (ablation) | MAN · CIFAR10 multi-target | attack success rate (%): 99.14 -> 99.50 | Improved |
| use concatenation label fusion (MANc) instead of recalibration (MANr) (ablation) | MAN · ImageNet multi-target, Res152 to Res101 | attack success rate (%): 49.69 -> 55.39 | Improved |
| condition generator on target label (MANc) versus single-target GAP | MANc · ImageNet VGG16 to VGG19 transfer | attack success rate (%): 70.83 -> 85.79 | Improved |
| fine-tune with MANc adversarial samples instead of GAP10 samples | MANc · CIFAR10 MI-FGSM strength 16sqrt(N) | classification accuracy (%): 79.08 -> 81.14 | Improved |
| fine-tune with adversarial samples from ten GAP models instead of one (ablation) | GAP · CIFAR10 MI-FGSM strength 16sqrt(N) | classification accuracy (%): 63.40 -> 79.08 | Improved |
| fine-tune ResNet32 with MANc adversarial samples | ResNet32 · CIFAR10 MI-FGSM strength 16sqrt(N) | classification accuracy (%): 10.58 -> 81.14 | Improved |
| increase perturbation threshold epsilon from very small values (ablation) | MAN · CIFAR10 VGG16 | attack success rate | Improved |
| increase perturbation threshold epsilon beyond 12sqrt(N) (ablation) | MAN · CIFAR10 VGG16 | attack success rate | Hurt |
| use a larger reconstruction weight alpha (ablation) | MAN · CIFAR10 VGG16 | attack success rate | Mixed |
Closest papers by meaning
- CLIP-Guided Generative Networks for Transferable Targeted Adversarial Attacks (ECCV 2024 · similarity 0.87)
- Any Target Can be Offense: Adversarial Example Generation via Generalized Latent Infection (ECCV 2024 · similarity 0.85)
- Dynamic Generative Targeted Attacks With Pattern Injection (CVPR 2023 · similarity 0.85)
- Dual-Flow: Transferable Multi-Target, Instance-Agnostic Attacks via $\textit{In-the-wild}$ Cascading Flow Optimization (NeurIPS 2025 · similarity 0.85)
- Clarifying the Behavior and the Difficulty of Adversarial Training (AAAI 2024 · similarity 0.83)
- Generate Universal Adversarial Perturbations for Few-Shot Learning (NeurIPS 2024 · similarity 0.83)
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.
