Research map / Federated learning and privacy / Privacy attacks and secure learning
A General Framework for Auditing Differentially Private Machine Learning
Fred Lu, Joseph Munoz, Maya Fuchs, Tyler LeBlond, Elliott V. Zaresky-Williams, Edward Raff, Francis Ferraro, Brian Testa
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 |
|---|---|---|---|
| influence-based perturbation attack instead of ClipBKD | ML-Audit · logistic regression, Naive Bayes, random forest | detected privacy lower bound | Improved |
| influence-based attack instead of Swap-X baseline | ML-Audit · iris, breast-cancer, output-perturbed logistic regression | detected privacy lower bound | No clear effect |
| target class prior (flip label) instead of only class mean (ablation) | Naive Bayes attack · Naive Bayes | detected privacy lower bound | Improved |
| use Katz-log interval instead of Clopper-Pearson (ablation) | ML-Audit · coverage simulation | confidence interval tightness | Improved |
| increase number of perturbed copies k at small eps (ablation) | ML-Audit · auditing at small eps | detected privacy lower bound | Improved |
| increase number of perturbed copies k at large eps (ablation) | ML-Audit · auditing at large eps | detected privacy lower bound | Hurt |
| add exposed class counts to summary tau | ML-Audit · Naive Bayes (diffprivlib) | detected privacy lower bound | Improved |
| add zero-vector and test predictions to tau (ablation) | ClipBKD · DP-SGD FMNIST | detected privacy lower bound | Improved |
| normalize FMNIST to [-1,1] or [0,1] instead of [0,255] (ablation) | DP-SGD audit · FMNIST | detected privacy risk: 0.7 -> 0.4 | Hurt |
| shadow adversarial attack instead of ClipBKD (ablation) | DP-SGD audit · FMNIST vs CIFAR10 | detected privacy lower bound | Mixed |
Closest papers by meaning
- Auditing Private Prediction (ICML 2024 · similarity 0.92)
- How Well Can Differential Privacy Be Audited in One Run? (NeurIPS 2025 · similarity 0.90)
- Auditing $f$-differential privacy in one run (ICML 2025 · similarity 0.89)
- Nearly Tight Black-Box Auditing of Differentially Private Machine Learning (NeurIPS 2024 · similarity 0.89)
- Position: Considerations for Differentially Private Learning with Large-Scale Public Pretraining (ICML 2024 · similarity 0.87)
- Attack-Aware Noise Calibration for Differential Privacy (NeurIPS 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.
