Research map / Federated learning and privacy / Privacy attacks and secure learning
Membership Privacy for Machine Learning Models Through Knowledge Transfer
Read the paperCheck your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's title only (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 |
|---|---|---|---|
| Replace direct privacy regularization with distillation onto selected reference data | DMP versus AdvReg · CIFAR100, DenseNet-12 | test accuracy at similar MIA risk: +138.1% | Helped |
| Replace direct privacy regularization with distillation onto selected reference data | DMP versus AdvReg · Purchase, Texas, CIFAR100, and CIFAR10 | test accuracy at similar MIA risk: +31.2%; +45.1%; +81.2%; +92.6%; +21.7% | Helped |
| Replace DP-SGD with DMP | CIFAR10 AlexNet · CIFAR10 | test accuracy at low whitebox MIA risk: 12.8% absolute increase | Helped |
| Select reference samples with lower-entropy teacher predictions | DMP · Purchase100 | whitebox MIA risk | Helped |
| Increase the teacher softmax temperature (ablation) | DMP · CIFAR100 AlexNet | whitebox MIA accuracy: 57.4 -> 55.6 -> 52.5 -> 51.7 | Helped |
| Increase the teacher softmax temperature (ablation) | DMP · CIFAR100 AlexNet | test accuracy: 37.3 -> 35.7 -> 32.8 -> 12.3 | Hurt |
| Reduce the reference-set size (ablation) | DMP · Purchase100 | membership inference risk | Helped |
| Reduce the reference-set size (ablation) | DMP · Purchase100 | test accuracy | Hurt |
| Increase GAN-generated reference-set size from 12.5k to 37.5k (ablation) | DMP · CIFAR10 | test accuracy: 53.0 -> 56.8 -> 57.5 | Helped |
| Increase GAN-generated reference-set size from 12.5k to 37.5k (ablation) | DMP · CIFAR10 | whitebox MIA accuracy: 50.3 -> 51.3 -> 52.1 | Hurt |
| Use synthetic rather than real reference data (ablation) | DMP · CIFAR10 | test accuracy: 53.0; 56.8; 57.5 | Mixed |
| Divide private training data among multiple teachers and aggregate their labels | PATE versus DMP · CIFAR10 | test accuracy: 68.1 vs 76.79% | Hurt |
Closest papers by meaning
- Impact of Dataset Properties on Membership Inference Vulnerability of Deep Transfer Learning (NeurIPS 2025 · similarity 0.88)
- DF-MIA: A Distribution-Free Membership Inference Attack on Fine-Tuned Large Language Models (AAAI 2025 · similarity 0.87)
- Mitigating Privacy Risk in Membership Inference by Convex-Concave Loss (ICML 2024 · similarity 0.86)
- Assessing Per-Sample Membership Inference Vulnerability without Retraining (NeurIPS 2026 · similarity 0.86)
- Membership Inference Attacks against Fine-tuned Large Language Models via Self-prompt Calibration (NeurIPS 2024 · similarity 0.86)
- Chameleon: Increasing Label-Only Membership Leakage with Adaptive Poisoning (ICLR 2024 · 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.