Research map / Federated learning and privacy / Differential privacy
Auditing Differentially Private Machine Learning: How Private is Private SGD?
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 |
|---|---|---|---|
| Use ClipBKD instead of membership inference | DP-SGD · Purchase-100, 2-layer neural network, εth=4 | εLB: 0.04 -> 0.46 | Helped |
| Use ClipBKD instead of standard backdoor poisoning | DP-SGD · Fashion-MNIST | εLB: 3.84x | Helped |
| Use ClipBKD instead of standard backdoor poisoning | DP-SGD · CIFAR10 | εLB | Helped |
| Increase the theoretical privacy parameter εth (reduce noise) (ablation) | DP-SGD audited with ClipBKD · Fashion-MNIST and Purchase-100, FNN | εLB | Helped |
| Reduce initialization randomness (ablation) | DP-SGD audited with ClipBKD · Fashion-MNIST, FNN | εLB | Helped |
| Reduce clipping norm (ablation) | DP-SGD audited with ClipBKD · Fashion-MNIST, FNN, non-fixed initialization | εLB | Hurt |
| Vary initialization randomness and clipping norm (ablation) | DP-SGD audited with ClipBKD · Purchase-100, FNN | εLB | No clear effect |
| Use fixed initialization instead of random initialization (ablation) | DP-SGD audited with ClipBKD · Fashion-MNIST, FNN, εth=∞ | εLB: 0.29 / 0.95 / 2.36 -> 4.54 / 4.54 / 4.54 | Helped |
Closest papers by meaning
- Auditing Private Prediction (ICML 2024 · similarity 0.91)
- Nearly Tight Black-Box Auditing of Differentially Private Machine Learning (NeurIPS 2024 · similarity 0.90)
- How Well Can Differential Privacy Be Audited in One Run? (NeurIPS 2025 · similarity 0.89)
- Auditing $f$-differential privacy in one run (ICML 2025 · similarity 0.89)
- The Last Iterate Advantage: Empirical Auditing and Principled Heuristic Analysis of Differentially Private SGD (ICLR 2025 · similarity 0.88)
- Beyond the Calibration Point: Mechanism Comparison in Differential Privacy (ICML 2024 · similarity 0.87)
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.