Research map / Federated learning and privacy / Differential privacy
Tight and Robust Private Mean Estimation with Few Users
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 the proposed user-level private mean-estimation algorithm | Mean estimation from n users with m samples each · High-dimensional mean estimation, user-level | ℓ2 error: min(O(r · √(d/m)), O(r · √d · log(d/δ) √log(1/δ) · √log(dn/α)/(ε · n · √m))) | Helped |
| Use the proposed user-level algorithm | Learning a discrete distribution from n users with m samples each · Discrete distribution learning over [d] | total variation distance: Õ(d/(ε · n · √m) + √(d/(m · n))) | Helped |
| Use an item-level private estimator requiring only O((1/ε) log(1/δ)) points | Private location estimation in ℝ^d · Item-level mean/location estimation | estimation error: O(r√d) | Helped |
| Combine the few-point estimator with Fast Johnson-Lindenstrauss rotation and block projections | Private item-level estimation in the intermediate-user regime · Item-level mean/location estimation, intermediate n | estimation error: O(r · √d · (1/ε) · log(d/δ) √log(1/δ) · √log(dn/α)/n) | Helped |
| Reduce each user's samples to their sample mean, then apply the item-level private algorithm | User-level mean estimation · High-dimensional mean estimation, user-level | ℓ2 error: O(r · √(d/m)) | Helped |
| Use the user-level method for discrete distribution learning | Distribution learning over [d] · Discrete distribution learning over [d], user-level | total variation distance: Õ(√(d/(mn) + d^2/(mn^2 ε^2))) | Helped |
| Use the proposed algorithm when up to 49% of users are adversarially corrupted | Robust user-level mean estimation · Mean estimation with corrupted users | robustness to corrupted users: 49% | Helped |
Closest papers by meaning
- Learning with User-Level Privacy (NeurIPS 2021 · similarity 0.92)
- Instance-optimal Mean Estimation Under Differential Privacy (NeurIPS 2021 · similarity 0.90)
- Optimal Survey Design for Private Mean Estimation (ICML 2025 · similarity 0.89)
- A Huber Loss Minimization Approach to Mean Estimation under User-level Differential Privacy (NeurIPS 2024 · similarity 0.88)
- Improved Communication-Privacy Trade-offs in $L_2$ Mean Estimation under Streaming Differential Privacy (ICML 2024 · similarity 0.88)
- Mean Estimation in the Add-Remove Model of Differential Privacy (ICML 2024 · 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.