Research map / Adversarial robustness and security / Backdoor and poisoning
Improving the Sensitivity of Backdoor Detectors via Class Subspace Orthogonalization
Guangmingmei Yang, David J. Miller, George Kesidis
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 CSO | MMBD · CIFAR-10 all-to-one LC | detection accuracy (%): 30 -> 76 | Improved |
| add CSO | MMBD · CIFAR-10 all-to-one Bpp | detection accuracy (%): 0 -> 36 | Improved |
| add CSO | MMBD · CIFAR-10 all-to-one BadNet | detection accuracy (%): 82 -> 92 | Improved |
| add CSO | PT-RED · CIFAR-10 all-to-one BadNet | detection accuracy (%): 0 -> 54 | Improved |
| add CSO | NC · CIFAR-10 all-to-one IAD | detection accuracy (%): 44 -> 74 | Improved |
| add CSO | MMBD · GTSRB all-to-one BadNet | detection accuracy (%): 52 -> 78 | Improved |
| add CSO | MMBD · CIFAR-10 all-to-one WaNet | detection accuracy (%): 90 -> 88 | No clear effect |
| split features at last conv layer instead of 9th (ablation) | MMBD-CSO · CIFAR-10 ML-BadNet | detection accuracy (%): 56 -> 70 | Improved |
| use class-specific feature masking instead of global mask (ablation) | MMBD-CSO · CIFAR-10 ML-BadNet | detection accuracy (%): 36 -> 70 | Improved |
| add ReLU to cosine similarity in CSO penalty (ablation) | MMBD-CSO · CIFAR-10 BadNet | detection accuracy (%): 70 -> 96 | Improved |
| raise CSO penalty lambda from 100 to 400 (ablation) | MMBD-CSO · CIFAR-10 ML-BadNet | detection accuracy (%): 44 -> 70 | Improved |
| use 10 clean images per class instead of 1 (ablation) | MMBD-CSO · CIFAR-10 ML-BadNet | detection accuracy (%): 44 -> 70 | Improved |
Closest papers by meaning
- Clean Data Can Still Carry Backdoors: Support-Persistent Backdoors for Model Reuse (NeurIPS 2026 · similarity 0.88)
- Lie Detector: Unified Backdoor Detection via Cross-Examination Framework (NeurIPS 2025 · similarity 0.87)
- Data Free Backdoor Attacks (NeurIPS 2024 · similarity 0.87)
- RPP: A Certified Poisoned-Sample Detection Framework for Backdoor Attacks under Dataset Imbalance (NeurIPS 2026 · similarity 0.87)
- Inspecting Prediction Confidence for Detecting Black-Box Backdoor Attacks (AAAI 2024 · similarity 0.86)
- Influence-Based Fair Selection for Sample-Discriminative Backdoor Attack (AAAI 2025 · 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.
