Research map / Adversarial robustness and security / Backdoor and poisoning
Label Poisoning is All You Need
Rishi Dev Jha, Jonathan Hayase, Sewoong Oh
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 |
|---|---|---|---|
| Increase flipped labels from 0 to 1500 | FLIP · CIFAR-10 ResNet-32 sinusoidal | poison test accuracy PTA (%): 00.1 -> 99.8 | Improved |
| Increase flipped labels from 0 to 1500 | FLIP · CIFAR-10 ResNet-32 sinusoidal | clean test accuracy CTA (%): 92.38 -> 89.87 | Hurt |
| Use pixel trigger instead of sinusoidal | FLIP · CIFAR-10 ResNet-32, 1500 labels | poison test accuracy PTA (%): 99.8 -> 29.9 | Hurt |
| Use FLIP instead of inner product baseline | inner product · CIFAR-10 ResNet-32 sinusoidal, 1500 labels | poison test accuracy PTA (%): 32.0 -> 99.8 | Improved |
| Use FLIP instead of random selection baseline | random selection · CIFAR-10 ResNet-32 sinusoidal, 1500 labels | poison test accuracy PTA (%): 23.0 -> 99.8 | Improved |
| Flip 5000 labels with inner product baseline to reach high PTA | inner product · CIFAR-10 ResNet-32 | CTA: 92.38 -> 82.84 | Hurt |
| Mismatch expert ResNet-32 and user ResNet-18 (ablation) | FLIP · CIFAR-10 sinusoidal, 1500 labels | poison test accuracy PTA (%): 99.8 -> 98.0 | No clear effect |
| Mismatch expert ResNet-32 and user VGG (ablation) | FLIP · CIFAR-10 sinusoidal, 1500 labels | poison test accuracy PTA (%): 99.8 -> 63.0 | Hurt |
| User trains with Adam instead of SGD (ablation) | FLIP r32 · CIFAR-10 sinusoidal, 1500 labels | poison test accuracy PTA (%): 99.8 -> 99.0 | No clear effect |
| Mismatch both architecture and optimizer (r32 -> r18) (ablation) | FLIP · CIFAR-10 sinusoidal, 1000 labels | poison test accuracy PTA (%): 86.9 -> 99.2 | Improved |
| Use traditionally poisoned teacher for distillation | student model · CIFAR-10 | poison test accuracy PTA (%): 99.9 -> 0.20 | No clear effect |
| Train expert models for K=0 epochs (ablation) | FLIP · CIFAR-10 sinusoidal | poison test accuracy PTA (%) | Hurt |
Closest papers by meaning
- Wicked Oddities: Selectively Poisoning for Effective Clean-Label Backdoor Attacks (ICLR 2025 · similarity 0.90)
- COMBAT: Alternated Training for Effective Clean-Label Backdoor Attacks (AAAI 2024 · similarity 0.90)
- A Set of Generalized Components to Achieve Effective Poison-only Clean-label Backdoor Attacks with Collaborative Sample Selection and Triggers (NeurIPS 2025 · similarity 0.88)
- Clean Data Can Still Carry Backdoors: Support-Persistent Backdoors for Model Reuse (NeurIPS 2026 · similarity 0.87)
- A Theoretical Analysis of Backdoor Poisoning Attacks in Convolutional Neural Networks (ICML 2024 · similarity 0.87)
- Clean-Label Graph Backdoor Attack in the Node Classification Task (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.
