Research map / Adversarial robustness and security / Attacks on text and graphs
Defending and Harnessing the Bit-Flip Based Adversarial Weight Attack
Zhezhi He, Adnan Siraj Rakin, Jingtao Li, Chaitali Chakrabarti, Deliang Fan
Check 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 |
|---|---|---|---|
| Apply piece-wise clustering to 8-bit weights | ResNet-20 · CIFAR-10 | NBF: 28.0±4.47 -> 58.79±4.14 | Improved |
| Use binarization-aware training | ResNet-20 · CIFAR-10 | NBF: 28.0±4.47 -> 541.2 ± 49.8 | Improved |
| Use binarization-aware training | ResNet-20 · CIFAR-10 | prior-attack accuracy: 91.84 -> 88.36 | Hurt |
| Apply Lasso-based pruning | ResNet-20 · CIFAR-10 | NBF: 28.0±4.47 -> 6.8±0.44 | Hurt |
| Apply adversarial weight training | ResNet-20 · CIFAR-10 | NBF: 28.0±4.47 -> 9.6±6.58 | Hurt |
| Remove batch normalization (ablation) | VGG-11 · CIFAR-10 | NBF: 16.4 ± 1.14 -> 9.2 ± 0.42 | Hurt |
| Increase dropout rate to 0.7 (ablation) | VGG-11 · CIFAR-10 | NBF: 16.4 ± 1.14 -> 24.6± 3.71 | Improved |
| Remove batch normalization (ablation) | VGG-11 with piece-wise clustering · CIFAR-10 | NBF: 29.79 ± 11.3 -> 8.8 ± 0.44 | Hurt |
| Increase dropout rate to 0.7 (ablation) | VGG-11 with piece-wise clustering · CIFAR-10 | NBF: 29.79 ± 11.3 -> 32.79 ± 3.49 | Improved |
| Increase network width to 2× (ablation) | ResNet-20 · CIFAR-10 | NBF: 28.0 ± 4.47 -> 26.2 ± 2.68 | No clear effect |
| Increase network width to 4× (ablation) | ResNet-20 · CIFAR-10 | NBF: 28.0 ± 4.47 -> 36.4 ± 12.44 | Improved |
| Increase network width to 4× (ablation) | ResNet-20 with piece-wise clustering · CIFAR-10 | NBF: 58.79 ± 4.14 -> 72 ± 18.79 | Improved |
Closest papers by meaning
- One-bit Flip is All You Need: When Bit-flip Attack Meets Model Training (ICCV 2023 · similarity 0.92)
- T-BFA: T argeted B it- F lip Adversarial Weight A ttack (TPAMI 2021 · similarity 0.92)
- Bit-Flip Attack: Crushing Neural Network With Progressive Bit Search (ICCV 2019 · similarity 0.90)
- Crossfire: An Elastic Defense Framework for Graph Neural Networks Under Bit Flip Attacks (AAAI 2025 · similarity 0.87)
- Your Scale Factors are My Weapon: Targeted Bit-Flip Attacks on Vision Transformers via Scale Factor Manipulation (CVPR 2025 · similarity 0.87)
- Nearest is Not Dearest: Towards Practical Defense against Quantization-conditioned Backdoor Attacks (CVPR 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.
