Research map / Adversarial robustness and security / Attacks on text and graphs
Amplifying Membership Exposure via Data Poisoning
Yufei Chen, Chao Shen, Yun Shen, Cong Wang, Yang Zhang
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 |
|---|---|---|---|
| Flip labels of poison samples from the target class | InceptionV3 classifier · CIFAR-10 airplane | MI AUC: 0.6917 -> 0.9255 | Improved |
| Add imperceptible, correctly labeled poisons optimized to match target-class features | Xception classifier · STL-10 | MI AUC: 0.578 -> 0.870 | Improved |
| Add imperceptible, correctly labeled poisons optimized to match target-class features | Xception classifier · STL-10 | Test accuracy: 0.861 -> 0.837 | Hurt |
| Reduce dirty-label poisoning budget to 100 (ablation) | ResNet50 classifier · CIFAR-10 | MI AUC: 0.5971 to 0.7076 | Improved |
| Use a shadow dataset from a different distribution (STL-10 rather than CIFAR-10) (ablation) | Dirty-label poisoning · CIFAR-10 airplane and cat | MI AUC: .058; .066; .057; .019; .019; .065; .045; -.009; -.017; .000 | Mixed |
| Fine-tune the feature extractor (ablation) | Dirty-label poisoning · CIFAR-10 | MI AUC | Improved |
| Fine-tune the feature extractor (ablation) | Clean-label poisoning · CIFAR-10 | MI AUC | No clear effect |
| Add L2 regularization during training (ablation) | Poisoned models · CIFAR-10 | MI AUC | Improved |
| Use early stopping during training (ablation) | Poisoned models · CIFAR-10 | MI AUC | Improved |
| Use DP-SGD during training (ablation) | Poisoned models · CIFAR-10 | MI AUC | Improved |
Closest papers by meaning
- Chameleon: Increasing Label-Only Membership Leakage with Adaptive Poisoning (ICLR 2024 · similarity 0.89)
- Wicked Oddities: Selectively Poisoning for Effective Clean-Label Backdoor Attacks (ICLR 2025 · similarity 0.86)
- Privacy Backdoors: Enhancing Membership Inference through Poisoning Pre-trained Models (NeurIPS 2024 · similarity 0.86)
- Assessing Per-Sample Membership Inference Vulnerability without Retraining (NeurIPS 2026 · similarity 0.86)
- Provable Watermarking for Data Poisoning Attacks (NeurIPS 2025 · similarity 0.86)
- Activation Gradient based Poisoned Sample Detection Against Backdoor Attacks (ICLR 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.
