Research map / Federated learning and privacy / Differential privacy
Private Estimation with Public Data
Alex Bie, Gautam Kamath, Vikrant Singhal
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 d+1 public samples to precondition private Gaussian data | Private Gaussian estimation under pure DP or zCDP · d-dimensional Gaussian, same distribution | dependence on distribution range parameters in private sample complexity: d+1 | Improved |
| Use one public sample as a coarse mean estimate | Private Gaussian mean estimation · Identity covariance Gaussian | dependence on mean range bound R: 1 | Improved |
| Allow public samples to come from a Gaussian within TV distance γ of the private distribution | Private Gaussian estimation · d-dimensional Gaussian | private sample complexity dependence on range parameters: d+1 | Improved |
| Increase public mixture samples from dimension-independent to dimension-linear in d | Private Gaussian mixture estimation · Separated Gaussian mixtures, approximate DP | private sample complexity: \frac{d^{1.5}k^{2.5}}{\eps} removed | Improved |
| Increase public mixture samples from dimension-independent to dimension-linear in d | Private Gaussian mixture estimation · Separated Gaussian mixtures, zCDP | dependence on distribution range parameters: m = O\left(\frac{d\log(k/\beta)}{w_{\min}}\right) | Improved |
| Use one public sample to set the mean range bound for CoinPress | CoinPress · Gaussian mean estimation, d=50, ρ=0.5, weak a priori bound (k=1000) | estimation error | Improved |
Closest papers by meaning
- Sample-Efficient Private Learning of Mixtures of Gaussians (NeurIPS 2024 · similarity 0.91)
- Public-data Assisted Private Stochastic Optimization: Power and Limitations (NeurIPS 2024 · similarity 0.89)
- Locally Optimal Private Sampling: Beyond the Global Minimax (NeurIPS 2025 · similarity 0.87)
- Dimension-free Private Mean Estimation for Anisotropic Distributions (NeurIPS 2024 · similarity 0.87)
- Privately Learning Smooth Distributions on the Hypercube by Projections (ICML 2024 · similarity 0.87)
- Instance-Optimality for Private KL Distribution Estimation (NeurIPS 2025 · 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.
