Research map / Federated learning and privacy / Differential privacy
Privately Publishable Per-instance Privacy
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 |
|---|---|---|---|
| Increase objective-perturbation regularization (lambda inflation) (ablation) | Obj-Pert logistic regression · Synthetic dataset (n = 1000, d = 50) | 0-1 loss | No clear effect |
| Use a data-dependent privacy report instead of the data-independent bound | Obj-Pert · UCI kidney dataset; logistic regression | accuracy of the pDP upper-bound estimate | Helped |
| Use a data-dependent privacy report instead of the data-independent bound | Obj-Pert · UCI wine quality dataset; linear regression | accuracy of the pDP upper-bound estimate | Helped |
| Allocate more of the fixed privacy budget to releasing the gradient and Hessian estimates, rather than nearly all of it to model release (ablation) | Data-dependent privacy report · Credit card default dataset | accuracy of the pDP upper-bound estimate | Helped |
| Reserve little privacy budget for the gradient and Hessian report while spending nearly all of it on model release (ablation) | Data-dependent privacy report · UCI credit default dataset | accuracy of the pDP upper-bound estimate | Hurt |
| Use a data-independent bound for GLM pDP reporting | Obj-Pert · Generalized linear models | dimension and dataset-size dependence of the bound | Helped |
| Use Gaussian Orthogonal Ensemble noise for releasing the Hessian instead of classical AnalyzeGauss noise | Hessian release mechanism · Hessian release | constant factor: roughly a constant factor of 2 | Helped |
| Use the data-independent GLM bound rather than directly exposing the ex-post pDP loss | Obj-Pert privacy reporting · Generalized linear models | additional privacy cost: no additional privacy budget | Helped |
| Release data-dependent gradient and Hessian estimates to tighten the report | Obj-Pert privacy reporting · Logistic regression on the UCI kidney dataset | accuracy of the pDP upper-bound estimate | Helped |
| Use the adaptive privacy report based on the released minimum Hessian eigenvalue | Obj-Pert privacy reporting · Well-conditioned Hessian | multiplicative approximation factor for the leverage-score term: from 12 to 1 + o(1) | Helped |
Closest papers by meaning
- Provable Privacy with Non-Private Pre-Processing (ICML 2024 · similarity 0.88)
- Instance-Specific Asymmetric Sensitivity in Differential Privacy (NeurIPS 2024 · similarity 0.88)
- Setting $\varepsilon$ is not the Issue in Differential Privacy (NeurIPS 2025 · similarity 0.87)
- Perturb-and-Project: Differentially Private Similarities and Marginals (ICML 2024 · similarity 0.87)
- One-shot Empirical Privacy Estimation for Federated Learning (ICLR 2024 · similarity 0.87)
- Prior-itizing Privacy: A Bayesian Approach to Setting the Privacy Budget in Differential Privacy (NeurIPS 2024 · 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.