Research map / Federated learning and privacy / Differential privacy
Efficient privacy loss accounting for subsampling and random allocation
Vitaly Feldman, Moshe Shenfeld
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 the PLD-based random allocation accounting instead of prior analytic bounds (hockey-stick and Renyi based) | 1-out-of-t random allocation with Gaussian mechanism · Gaussian mechanism, delta = 10^-6, varying sigma and t | privacy parameter epsilon | Improved |
| compare PLD-based bound to Monte Carlo estimates | random allocation privacy profile · Chua et al. 2024a setting | privacy profile | Improved |
| use random allocation instead of Poisson sampling at same noise and expected batch size (Gaussian) | DP-SGD privacy accounting · nearly all parameter regimes | privacy profile | Improved |
| use random allocation instead of Poisson sampling when epsilon is much smaller than 1 | Gaussian mechanism · epsilon << 1 | privacy profile | Mixed |
| use random allocation instead of Poisson sampling with Laplace mechanism | Laplace mechanism · varying sigma and t | privacy guarantee | Mixed |
| use PLD accounting instead of RDP-based accounting for DP-SGD with PREAMBLE | PREAMBLE (n=6e5, d=2^20, C=2^15, E=10) · (epsilon=1, delta=10^-6) | noise ratio sigma vs plain Gaussian DP-SGD | Improved |
| use random allocation instead of Poisson subsampling in synthetic Bernoulli mean estimation | noisy sum estimation · Feldman & Shenfeld 2025 toy example | accuracy at fixed privacy | Improved |
| decrease discretization parameter alpha (ablation) | PLD random allocation algorithm · runtime on laptop | runtime | Hurt |
Closest papers by meaning
- Privacy amplification by random allocation (NeurIPS 2025 · similarity 0.94)
- Sampling-Free Privacy Accounting for Matrix Mechanisms under Random Allocation (NeurIPS 2026 · similarity 0.91)
- A Generalized Shuffle Framework for Privacy Amplification: Strengthening Privacy Guarantees and Enhancing Utility (AAAI 2024 · similarity 0.90)
- Individualized Privacy Accounting via Subsampling with Applications in Combinatorial Optimization (ICML 2024 · similarity 0.90)
- Optimal Survey Design for Private Mean Estimation (ICML 2025 · similarity 0.89)
- Scalable DP-SGD: Shuffling vs. Poisson Subsampling (NeurIPS 2024 · similarity 0.89)
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.
