Research map / Adversarial robustness and security / Certified and empirical robustness
GradMDM: Adversarial Attack on Dynamic Networks
Jianhong Pan, Lin Geng Foo, Qichen Zheng, Zhipeng Fan, Hossein Rahmani, Qiuhong Ke, Jun 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 Power Loss and Complexity Gradient Masking (GradMDM vs ILFO) | ILFO · ImageNet SkipNet+HRL | ARP (%): 49.6 -> 73.4 | Improved |
| GradMDM vs ILFO | ILFO · CIFAR-10 SkipNet-HRL | ARP (%): 5.2 -> 30.0 | Improved |
| GradMDM vs ILFO on dynamic width network | ILFO · ManiDP ImageNet | ARP (%): 85.6 -> 99.3 | Improved |
| GradMDM vs ILFO on transformers and early-exit CNN | ILFO · DVT ImageNet | ARP (%): 85.0 -> 98.3 | Improved |
| add Complexity Gradient Masking (ablation) | Baseline Loss · ImageNet SkipNet+HRL | ARP (%): 49.6 -> 54.0 | Improved |
| add Power Loss (ablation) | Baseline Loss · ImageNet SkipNet+HRL | ARP (%): 49.6 -> 69.0 | Improved |
| combine Power Loss with CGM (ablation) | Power Loss · ImageNet SkipNet+HRL | ARP (%): 69.0 -> 73.4 | Improved |
| increase gamma (imperceptibility weight) from 1e+0 to 1e+4 (ablation) | GradMDM · ImageNet SkipNet+HRL | ARP (%) and PSNR: 76.1 -> 57.5 ARP, 54.3 -> 73.9 PSNR | Mixed |
| increase Power Loss exponent alpha from 1 to 4 (ablation) | GradMDM · ImageNet SkipNet+HRL | ARP (%): 50.7 -> 73.4 | Improved |
| increase alpha beyond 4 (ablation) | GradMDM · ImageNet SkipNet+HRL | ARP (%): 73.4 -> 72.5 (alpha 5), 69.0 (alpha 6) | No clear effect |
| update with Finished Gradient jointly instead of masking (ablation) | GradMDM · ImageNet SkipNet+HRL | ARP (%): 73.4 -> 38.1 | Hurt |
| project out conflicting direction via CGM instead of adding L_F gradient (ablation) | GradMDM · ImageNet SkipNet+HRL | ARP (%): 66.8 -> 73.4 | Improved |
Closest papers by meaning
- Towards Effective and Sparse Adversarial Attack on Spiking Neural Networks via Breaking Invisible Surrogate Gradients (CVPR 2025 · similarity 0.86)
- Improving Generalization of Universal Adversarial Perturbation via Dynamic Maximin Optimization (AAAI 2025 · similarity 0.85)
- If You Want to Be Robust, Be Wary of Initialization (NeurIPS 2024 · similarity 0.85)
- Efficient Robustness Evaluation via Constraint Relaxation (AAAI 2025 · similarity 0.84)
- MemFreezing: A Novel Adversarial Attack on Temporal Graph Neural Networks under Limited Future Knowledge (ICML 2025 · similarity 0.84)
- Out of Thin Air: Exploring Data-Free Adversarial Robustness Distillation (AAAI 2024 · similarity 0.84)
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.
