Research map / Adversarial robustness and security / Backdoor and poisoning
An Invisible Black-Box Backdoor Attack through Frequency Domain
Tong Wang, Yuan Yao, Feng Xu, Shengwei An, Hanghang Tong, Ting Wang
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 |
|---|---|---|---|
| inject trigger in RGB channels instead of UV channels (ablation) | FTrojan (mix) · GTSRB | ASR: 99.25 -> 92.05 | Hurt |
| inject trigger in UV channels instead of RGB channels (ablation) | FTrojan (mix) · CIFAR10 | ASR: 95.99 -> 99.97 | Improved |
| use mixed mid and high frequency triggers instead of single band (ablation) | FTrojan (UV) · MNIST/GTSRB/CIFAR10/ImageNet/PubFig | ASR: UV+mid 99.22, UV+high 99.81, UV+mix 99.94 on MNIST | No clear effect |
| UV only instead of all YUV channels (ablation) | FTrojan (mix) · GTSRB | ASR: 98.35 -> 99.25 | No clear effect |
| raise injection rate from 0.01% to 1% or more (ablation) | FTrojan · GTSRB and CIFAR10 | ASR: high ASR when injection rate is no less than 1% | Improved |
| vary injection rate across wide range (ablation) | FTrojan · GTSRB and CIFAR10 | BA | No clear effect |
| raise trigger magnitude above a threshold (1 to 50 sweep) (ablation) | FTrojan · GTSRB and CIFAR10 | ASR | Improved |
| FTrojan vs BadNet, SIG, Refool, IAB trigger | backdoor attack · CIFAR10 | ASR: 99.97 vs 94.80 (BadNet), 95.76 (SIG), 73.20 (Refool), 79.70 (IAB) | Improved |
| FTrojan vs existing attacks on fidelity | backdoor attack · GTSRB | PSNR: 40.9 vs 24.9 (BadNet) | Improved |
| apply Neural Cleanse defense | FTrojan · CIFAR10 | abnormal index: 1.25 clean -> 1.85 backdoored | No clear effect |
| apply STRIP defense | FTrojan · CIFAR10 | false acceptance rate: 77.40% | No clear effect |
| apply Februus defense | FTrojan · CIFAR10 | BA/ASR: BA 86.42 -> 10.60, ASR 99.55 -> 76.73 | Mixed |
Closest papers by meaning
- Rethinking CNN’s Generalization to Backdoor Attack from Frequency Domain (ICLR 2024 · similarity 0.91)
- A Dual Stealthy Backdoor: From Both Spatial and Frequency Perspectives (AAAI 2024 · similarity 0.91)
- Inspecting Prediction Confidence for Detecting Black-Box Backdoor Attacks (AAAI 2024 · similarity 0.87)
- WaveAttack: Asymmetric Frequency Obfuscation-based Backdoor Attacks Against Deep Neural Networks (NeurIPS 2024 · similarity 0.87)
- Attention-Imperceptible Backdoor Attacks on Vision Transformers (AAAI 2025 · similarity 0.87)
- Invisible Backdoor Attack against Self-supervised Learning (CVPR 2025 · similarity 0.87)
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.
