Research map / Federated learning and privacy / Differential privacy
On Differentially Private Sampling from Gaussian and Product Distributions
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 |
|---|---|---|---|
| Generate a sample instead of learning the distribution | DP Gaussian methods with known covariance · Gaussian, known covariance | sample complexity: $\tilde{\Theta}\left(\frac{d}{\alpha^2} + \frac{d}{\alpha\epsilon}\right)$ for learning versus $\tilde{\Theta}\left(\fra | Helped |
| Generate a sample instead of learning the distribution | DP Gaussian methods with bounded covariance · Gaussian, bounded covariance | sample complexity: $\tilde{\Theta}\left(\frac{d^2}{\alpha^2} + \frac{d^2}{\alpha\epsilon}\right)$ for learning versus $\tilde{\Theta}\left( | Helped |
| Use a constant-accuracy private learner for preconditioning before sampling | DP Gaussian sampling with unknown unbounded covariance · Gaussian, unbounded covariance | dependence on accuracy parameter $\alpha$ in sample complexity: Sampling: $\tilde{O}\left(\frac{d^2}{\epsilon}\right)$; learning: $\tilde{\Theta}\left(\frac{d^2}{\alpha^2} + \frac{d^2} | Helped |
| Replace approximate-DP sampling with pure-DP sampling | Product-distribution samplers · Product distributions on $\{0, 1\}^d$ | privacy guarantee at comparable sample complexity: Both have sample complexity $\tilde{\Theta}\left(\frac{d}{\alpha\epsilon}\right)$ | Helped |
| Use the proposed pure-DP sampler rather than the previous approximate-DP sampler | Product-distribution samplers when $\alpha \leq 1/\log d$ · Product distributions on $\{0, 1\}^d$ | sample complexity: O\left(\frac{d \log(d/\alpha)}{\alpha\epsilon}\right) versus $O\left(\frac{d\sqrt{\log(1/\delta)}}{\alpha\epsilon}\left( | Helped |
Closest papers by meaning
- Differentially Private Sampling from Distributions (NeurIPS 2021 · similarity 0.92)
- Sample-Efficient Private Learning of Mixtures of Gaussians (NeurIPS 2024 · similarity 0.91)
- Privately Learning Smooth Distributions on the Hypercube by Projections (ICML 2024 · similarity 0.89)
- Locally Optimal Private Sampling: Beyond the Global Minimax (NeurIPS 2025 · similarity 0.88)
- Exactly Minimax-Optimal Locally Differentially Private Sampling (NeurIPS 2024 · similarity 0.88)
- Privacy Guarantees in Posterior Sampling under Contamination (NeurIPS 2026 · 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.