Research map / Adversarial robustness and security / Attacks on text and graphs
TBT: Targeted Neural Network Attack With Bit Trojan
Check your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's title only (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 |
|---|---|---|---|
| Insert a bit Trojan rather than use training-time Trojan insertion | VGG-16 · CIFAR-10 | number of weights changed: 150 vs 5120 and 11M | Helped |
| Insert a bit Trojan rather than use training-time Trojan insertion | VGG-16 · SVHN | ASR (%): 73.81 vs 75.50; BadNet 99.98 | Mixed |
| Increase trigger area from 6.25% to 11.82% (ablation) | TBT · CIFAR-10 | ASR (%): 89.40 -> 95.97 | Helped |
| Increase trigger area from 6.25% to 11.82% (ablation) | TBT · CIFAR-10 | number of bits flipped: 645 -> 627 | Mixed |
| Increase the number of modified weights from 10 to 188 (ablation) | TBT · CIFAR-10 | ASR (%): 79.70 -> 95.62 | Helped |
| Increase the number of modified weights from 10 to 188 (ablation) | TBT · CIFAR-10 | number of bits flipped: 37 -> 803 | Hurt |
| Place the trigger at top left rather than bottom right (ablation) | TBT, target class 3 · CIFAR-10 | ASR (%): 84.94 -> 96.44 | Helped |
| Place the trigger at center rather than bottom right (ablation) | TBT, target class 4 · CIFAR-10 | ASR (%): 89.55 -> 98.27 | Helped |
| Place the trigger at top left rather than bottom right (ablation) | TBT, target class 6 · CIFAR-10 | ASR (%): 80.82 -> 96.41 | Helped |
| Place the trigger at top left rather than bottom right (ablation) | TBT, target class 7 · CIFAR-10 | ASR (%): 85.40 -> 91.91 | Helped |
| Restrict trigger noise to 0.2% of the maximum pixel value (ablation) | TBT · CIFAR-10 | ASR (%): 79% | Hurt |
| Use a Network In Network topology without a fully connected output layer (ablation) | TBT · CIFAR-10 | ASR (%): 99 % | Mixed |
Closest papers by meaning
- Deep-TROJ: An Inference Stage Trojan Insertion Algorithm through Efficient Weight Replacement Attack (CVPR 2024 · similarity 0.91)
- Targeted Attack against Deep Neural Networks via Flipping Limited Weight Bits (ICLR 2021 · similarity 0.88)
- Your Scale Factors are My Weapon: Targeted Bit-Flip Attacks on Vision Transformers via Scale Factor Manipulation (CVPR 2025 · similarity 0.85)
- Scanning Trojaned Models Using Out-of-Distribution Samples (NeurIPS 2024 · similarity 0.84)
- TrojanDec: Data-free Detection of Trojan Inputs in Self-supervised Learning (AAAI 2025 · similarity 0.83)
- From Toxic to Trustworthy: Using Self-Distillation and Semi-supervised Methods to Refine Neural Networks (AAAI 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.