Research map / Federated learning and privacy / Differential privacy
Privacy Induces Robustness: Information-Computation Gaps and Sparse Mean Estimation
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 |
|---|---|---|---|
| Require pure DP and high-probability accuracy | Private mechanisms · Adversarially corrupted datasets | robust corruption fraction: \eta(n) = O \Paren{ \min \Paren{ \frac{\log 1/\beta}{\varepsilon n}, \frac{\log 1/\delta}{\varepsilon n + \log n} } } | Helped |
| Use the SoS exponential mechanism for sparse mean estimation | Sparse mean estimation · Gaussian sparse mean estimation | sample complexity: \frac{ k^2 + \log(1/\beta) + \log \log R }{\alpha^2\varepsilon} | Helped |
| Use a polynomial-time private algorithm with high success probability | Sparse mean estimation · Gaussian sparse mean estimation | sample complexity lower bound: n \geq \tfrac{k^2 m^2}{\alpha^4 \cdot \text{poly} \log(d,\tfrac{1}{\varepsilon},\tfrac{1}{\alpha})} | Hurt |
| Use Threshold rather than SoS | Private sparse mean estimation · Gaussian sparse mean estimation | sample complexity dependence on success probability: \frac{k^2 + \log(1/\beta)}{\alpha^2\varepsilon} -> \frac{k^2 \log(1/\beta)}{\alpha^2\varepsilon} | Hurt |
| Use polynomial time instead of exponential time | Private Gaussian mean estimation · Gaussian mean estimation | sample complexity: n \geq \min \{ \tfrac{\log(1/\beta)}{\varepsilon \alpha} \cdot (\log(1/\alpha))^{1/2 - o(1)}, \tfrac{2^{d^c}\log(1/\beta | Hurt |
| Require privacy for covariance testing | Gaussian covariance testing · Σ=I versus ‖Σ−I‖F ≥ 1/2 | sample complexity lower bound: n \geq \Omega(\min( C \log(1/\beta)/\varepsilon, C' d^2 )) | Hurt |
Closest papers by meaning
- From Robustness to Privacy and Back (ICML 2023 · similarity 0.93)
- Provable Robustness against Backdoor Attacks via the Primal-Dual Perspective on Differential Privacy (NeurIPS 2026 · similarity 0.89)
- Truthful High Dimensional Sparse Linear Regression (NeurIPS 2024 · similarity 0.88)
- Improved Communication-Privacy Trade-offs in $L_2$ Mean Estimation under Streaming Differential Privacy (ICML 2024 · similarity 0.88)
- Privacy Guarantees in Posterior Sampling under Contamination (NeurIPS 2026 · similarity 0.88)
- Universal Exact Compression of Differentially Private Mechanisms (NeurIPS 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.