Research map / Adversarial robustness and security / Backdoor and poisoning
ProFlip: Targeted Trojan Attack With Progressive Bit Flips
Huili Chen, Cheng Fu, Jishen Zhao, Farinaz Koushanfar
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 progressive critical bits search with parameter sensitivity instead of TBT's last-layer fine-tuning | ProFlip vs TBT · CIFAR-10 ResNet-18 | bit flips: 413 -> 12 | Improved |
| select the vulnerable parameter by fitness score (psens=45) instead of last-layer weights (psens=60) (ablation) | ProFlip BFA · SVHN VGG-16 | ASR: 73.8 -> 94.5 | Improved |
| increase trigger area from 6.25% to 11.82% (ablation) | ProFlip · CIFAR-10 ResNet-18 | bit flips: 42 -> 3 | Improved |
| increase trigger area from 6.25% to 11.82% (ablation) | ProFlip trigger generation · CIFAR-10 ResNet-18 | trigger ASR: 25.65 -> 66.3 | Improved |
| increase clean sample size from 64 to 256 images (ablation) | ProFlip · CIFAR-10 ResNet-18 | ASR: 95.7 -> 97.9 | Improved |
| increase clean sample size from 256 to 1024 (ablation) | ProFlip · CIFAR-10 ResNet-18 | ASR: 97.9 -> 97.9 | No clear effect |
| vary target class (class 6 vs class 2) (ablation) | ProFlip · CIFAR-10 ResNet-18 | bit flips: 12 -> 4 | Mixed |
| increase gamma1 from 1 to 8 in Trojan loss (ablation) | ProFlip · CIFAR-10 ResNet-18 | ASR / test accuracy: ASR 97.68 -> 96.31; TA 89.44 -> 90.38 | Mixed |
| decompose the most vulnerable parameter as a defense | ProFlip · CIFAR-10 ResNet-18 | ASR: 97.9 -> 73.7 | Hurt |
| decompose the most vulnerable parameter as a defense | ProFlip · CIFAR-10 ResNet-18 | bit flips: 12 -> 111 | Improved |
| decompose top-3 sensitive parameters of VGG-16 as a defense | ProFlip · SVHN VGG-16 | bit flips: 20 -> 124 | Improved |
Later papers that compared with this method
3 later accepted papers report ProFlip as a baseline in their results tables; 3 of them report a better number on most of the results they share with it.
- Backdoor Attacks on Neural Networks via One-Bit Flip (ICCV 2025 · better on 5 of 5 shared results)
- Your Scale Factors are My Weapon: Targeted Bit-Flip Attacks on Vision Transformers via Scale Factor Manipulation (CVPR 2025 · better on 4 of 4 shared results)
- TrojViT: Trojan Insertion in Vision Transformers (CVPR 2023 · better on 1 of 1 shared result)
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- TBT: Targeted Neural Network Attack With Bit Trojan (CVPR 2020 · similarity 0.90)
- Bit-Flip Attack: Crushing Neural Network With Progressive Bit Search (ICCV 2019 · similarity 0.87)
- Deep-TROJ: An Inference Stage Trojan Insertion Algorithm through Efficient Weight Replacement Attack (CVPR 2024 · similarity 0.87)
- Hardly Perceptible Trojan Attack against Neural Networks with Bit Flips (ECCV 2022 · similarity 0.86)
- Targeted Attack against Deep Neural Networks via Flipping Limited Weight Bits (ICLR 2021 · similarity 0.86)
- Practical Detection of Trojan Neural Networks: Data-Limited and Data-Free Cases (ECCV 2020 · similarity 0.86)
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.
