Research map / Adversarial robustness and security / Backdoor and poisoning
Hardly Perceptible Trojan Attack against Neural Networks with Bit Flips
Jiawang Bai, Kuofeng Gao, Dihong Gong, Shu-Tao Xia, Zhifeng Li, Wei 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 |
|---|---|---|---|
| jointly optimize bit flips, noise and flow field instead of two stages (ablation) | HPT · CIFAR-10 ResNet-18 | attack success rate: 93.2 -> 94.1 | Improved |
| jointly optimize instead of trigger-only without bit flips (ablation) | HPT · ImageNet ResNet-18 | attack success rate: 70.5 -> 95.2 | Improved |
| jointly optimize instead of two stage (ablation) | HPT · SVHN VGG-16 | attack success rate: 74.6 -> 78.0 | Improved |
| apply transparent trigger to TBT | TBT · CIFAR-10 ResNet-18 | attack success rate: 90.2 -> 56.6 | Hurt |
| apply transparent trigger to TrojanNN | TrojanNN · CIFAR-10 ResNet-18 | attack success rate: 93.9 -> 73.5 | Hurt |
| HPT instead of TBT | TBT · CIFAR-10 ResNet-18 | number of bit flips: 540 -> 12 | Improved |
| HPT instead of TBT | TBT · ImageNet ResNet-18 | attack success rate: 100.0 -> 95.2 | Hurt |
| increase noise bound epsilon from 0.01 to 0.04 (ablation) | HPT · CIFAR-10 ResNet-18 | attack success rate: 10.4 -> 94.1 | Improved |
| increase noise bound epsilon from 0.04 to 0.05 (ablation) | HPT · CIFAR-10 ResNet-18 | attack success rate: 94.1 -> 97.9 | Improved |
| increase flow smoothness bound kappa from 0.005 to 0.025 (ablation) | HPT · CIFAR-10 ResNet-18 | attack success rate: 93.0 -> 96.4 | Improved |
| add piece-wise clustering defense | HPT · CIFAR-10 ResNet-18 | attack success rate: 94.1 -> 86.6 | Hurt |
| raise bit budget b from 10 to 40 under defense (ablation) | HPT · CIFAR-10 ResNet-18 with defense | attack success rate: 86.6 -> 93.2 | Improved |
Closest papers by meaning
- TBT: Targeted Neural Network Attack With Bit Trojan (CVPR 2020 · similarity 0.92)
- Practical Detection of Trojan Neural Networks: Data-Limited and Data-Free Cases (ECCV 2020 · similarity 0.89)
- Deep-TROJ: An Inference Stage Trojan Insertion Algorithm through Efficient Weight Replacement Attack (CVPR 2024 · similarity 0.86)
- Your Scale Factors are My Weapon: Targeted Bit-Flip Attacks on Vision Transformers via Scale Factor Manipulation (CVPR 2025 · similarity 0.85)
- TrojanDec: Data-free Detection of Trojan Inputs in Self-supervised Learning (AAAI 2025 · similarity 0.83)
- A Dual Stealthy Backdoor: From Both Spatial and Frequency Perspectives (AAAI 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.
