Research map / Adversarial robustness and security / Attacks on text and graphs
GradAuto: Energy-Oriented Attack on Dynamic Neural Networks
Jianhong Pan, 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 Rectified Direction (gradient projection removing conflict between Complexity and Finished gradients) (ablation) | ILFO · ImageNet SkipNet+HRL | Complexity increase (%): 46.9 -> 52.3 | Improved |
| add Rectified Magnitude (Lipschitz-bounded gradient magnitude) (ablation) | ILFO · ImageNet SkipNet+HRL | Complexity increase (%): 46.9 -> 86.9 | Improved |
| add Rectified Direction on top of Rectified Magnitude (ablation) | Rectified Magnitude · ImageNet SkipNet+HRL | Complexity increase (%): 86.9 -> 88.9 | Improved |
| replace ILFO with GradAuto on SkipNet+HRL | ILFO · ImageNet SkipNet+HRL | FLOPs recovery (%): 46.9 -> 88.9 | Improved |
| replace ILFO with GradAuto on SkipNet, CIFAR-10 | ILFO · CIFAR-10 SkipNet | FLOPs recovery (%): 84.29 -> 100 | Improved |
| replace ILFO with GradAuto on SACT, CIFAR-10 | ILFO · CIFAR-10 SACT | FLOPs recovery (%): 72.49 -> 100 | Improved |
| replace ILFO with GradAuto on ManiDP, ImageNet | ILFO · ImageNet ManiDP | FLOPs recovery (%): 85.6 -> 100 | Improved |
| lower the Lipschitz constant from 1e+0 to 1e-3 (ablation) | GradAuto · ImageNet SkipNet+HRL | FLOPs recovery (%): 88.9 -> 71.0 | Mixed |
| raise Lipschitz constant from 1e-3 to 1e+0 (ablation) | GradAuto · ImageNet SkipNet+HRL | FLOPs recovery (%): 71.0 -> 88.9 | Mixed |
| raise Lipschitz constant beyond 1e+0 (to 1e+2) (ablation) | GradAuto · ImageNet SkipNet+HRL | FLOPs recovery (%): 88.9 -> 88.9 | No clear effect |
Closest papers by meaning
- GradMDM: Adversarial Attack on Dynamic Networks (TPAMI 2023 · similarity 0.92)
- Towards Effective and Sparse Adversarial Attack on Spiking Neural Networks via Breaking Invisible Surrogate Gradients (CVPR 2025 · similarity 0.84)
- ILFO: Adversarial Attack on Adaptive Neural Networks (CVPR 2020 · similarity 0.84)
- Exponential Dynamic Energy Network for High Capacity Sequence Memory (NeurIPS 2025 · similarity 0.82)
- Are Conventional SNNs Really Efficient? A Perspective from Network Quantization (CVPR 2024 · similarity 0.82)
- If You Want to Be Robust, Be Wary of Initialization (NeurIPS 2024 · 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.
