Research map / Federated learning and privacy / Differential privacy
A Randomized Approach to Tight Privacy Accounting
Jiachen T. Wang, Saeed Mahloujifar, Tong Wu, Ruoxi Jia, Prateek Mittal
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 |
|---|---|---|---|
| Use EVR with DP-SGD rather than relying directly on a strict upper bound | Fine-tuning ImageNet-pretrained BEiT · CIFAR100 | test error: around 7% (relative) error reduction when ε=0.6 | Improved |
| Use importance sampling for online privacy accounting | MC accountant · Online accounting, (σ, δ, q) = (1.0, 10^{-9}, 10^{-3}) | relative error | Improved |
| Reuse previously drawn samples for each new composition step | MC accountant · Online privacy accounting | runtime: around 5 times faster | Improved |
| Use GPU acceleration instead of CPU-only sampling | MC sampling of Gaussian mixtures · Gaussian mixture sampling | speedup: 10^3 times | Improved |
| Use a higher τ in EVR verification | EVR · Composed Gaussian mechanisms | privacy bound tightness | Improved |
| Use a higher τ in EVR verification | EVR · Composed Gaussian mechanisms | runtime: <2 minutes | Hurt |
| Replace simple Monte Carlo sampling with exponential-tilting importance sampling (ablation) | MC estimator for Poisson Subsampled Gaussian composition · Privacy verification | estimator second moment | Improved |
| Set θ to the proposed heuristic rather than θ=0 (ablation) | Importance Sampling estimator · Poisson Subsampled Gaussian composition | estimator second moment | Improved |
Closest papers by meaning
- Auditing Private Prediction (ICML 2024 · similarity 0.90)
- Provable Privacy with Non-Private Pre-Processing (ICML 2024 · similarity 0.90)
- Purifying Approximate Differential Privacy with Randomized Post-processing (NeurIPS 2025 · similarity 0.89)
- The Last Iterate Advantage: Empirical Auditing and Principled Heuristic Analysis of Differentially Private SGD (ICLR 2025 · similarity 0.89)
- 𝑓-Differential Privacy Filters: Validity and Approximate Solutions (NeurIPS 2026 · similarity 0.89)
- Auditing $f$-differential privacy in one run (ICML 2025 · similarity 0.88)
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.
