Research map / Adversarial robustness and security / Backdoor and poisoning
Randomized Channel Shuffling: Minimal-Overhead Backdoor Attack Detection without Clean Datasets
Ruisi Cai, Zhenyu Zhang, Tianlong Chen, Xiaohan Chen, Zhangyang 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 |
|---|---|---|---|
| Detect with randomized channel shuffling instead of Neural Cleanse | backdoor detection · BadNets CIFAR-10 ResNet-20 | AUROC: 0.669 -> 1.000 | Improved |
| Detect with randomized channel shuffling instead of STRIP | backdoor detection · BadNets CIFAR-10 ResNet-20 | AUROC: 0.459 -> 1.000 | Improved |
| Detect with channel shuffling instead of Neural Cleanse | backdoor detection · WaNet CIFAR-10 ResNet-18 | AUROC: 0.314 -> 0.954 | Improved |
| Detect with channel shuffling instead of STRIP | backdoor detection · BadNets GTSRB ResNet-20 | AUROC: 0.587 -> 0.880 | Improved |
| Move trigger to bottom left (ablation) | Ours · BadNets CIFAR-10 ResNet-20 | AUROC: 1.000 | No clear effect |
| Random trigger location (ablation) | Ours · BadNets CIFAR-10 ResNet-20 | AUROC: 1.000 | No clear effect |
| Use RGB trigger instead of gray-scale (ablation) | Ours · BadNets CIFAR-10 ResNet-20 | AUROC: 1.000 | No clear effect |
| Increase number of target classes from 2 to 5 (ablation) | Ours · BadNets CIFAR-10 ResNet-20 | AUROC: 0.978 -> 0.770 | Hurt |
| Source-label-specific (partial) backdoor (ablation) | Ours · BadNets CIFAR-10 ResNet-20 | AUROC: 0.945 | Mixed |
| Retrain with randomized channel shuffling | backdoored ResNet-20 · BadNets CIFAR-10 | test accuracy: 91.4 ± 0.5% -> 16.2 ± 7.3% | Hurt |
| Retrain with randomized channel shuffling | backdoored ResNet-20 · BadNets CIFAR-10 | attack success rate: 100.0 ± 0.0% -> 100.0 ± 0.0% | No clear effect |
| Retrain with randomized channel shuffling | backdoored AlexNet · BadNets CIFAR-10 | test accuracy: 85.7 ± 0.5% -> 45.4 ± 6.7% | Hurt |
Closest papers by meaning
- RPP: A Certified Poisoned-Sample Detection Framework for Backdoor Attacks under Dataset Imbalance (NeurIPS 2026 · similarity 0.90)
- BackdoorDM: A Comprehensive Benchmark for Backdoor Learning on Diffusion Model (NeurIPS 2025 · similarity 0.90)
- From Toxic to Trustworthy: Using Self-Distillation and Semi-supervised Methods to Refine Neural Networks (AAAI 2024 · similarity 0.89)
- BaDExpert: Extracting Backdoor Functionality for Accurate Backdoor Input Detection (ICLR 2024 · similarity 0.89)
- Lie Detector: Unified Backdoor Detection via Cross-Examination Framework (NeurIPS 2025 · similarity 0.88)
- Backdoor Defense via Test-Time Detecting and Repairing (CVPR 2024 · similarity 0.88)
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.
