Research map / Federated learning and privacy / Differential privacy
Improving the Privacy and Practicality of Objective Perturbation for Differentially Private Linear Learners
Rachel Emily Redberg, Antti Koskela, Yu-Xiang Wang
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 Algorithm 1 instead of honest DP-SGD | logistic regression, eps=0.1 · Synthetic-L | test accuracy (%): 77.34% -> 90.05% | Improved |
| Use Algorithm 1 instead of honest DP-SGD | logistic regression, eps=1 · Synthetic-L | test accuracy (%): 93.15% -> 94.50% | Improved |
| Use Algorithm 1 instead of honest DP-SGD | logistic regression, eps=8 · Synthetic-L | test accuracy (%): 95.17 % -> 95.30% | Improved |
| Use Algorithm 1 instead of honest DP-SGD | logistic regression, eps=0.1 · Adult | test accuracy (%): 78.32% -> 81.37% | Improved |
| Use Algorithm 1 instead of honest DP-SGD | logistic regression, eps=1 · Adult | test accuracy (%): 82.40% -> 83.18% | Improved |
| Use Algorithm 1 instead of honest DP-SGD | logistic regression, eps=8 · Adult | test accuracy (%): 83.66% -> 83.99% | Improved |
| Account for privacy cost of hyperparameter tuning (honest vs dishonest DP-SGD) | DP-SGD · Synthetic-L, eps=0.1 | test accuracy (%): 93.85% -> 77.34% | Hurt |
| Account for privacy cost of hyperparameter tuning (honest vs dishonest DP-SGD) | DP-SGD · Adult, eps=0.1 | test accuracy (%): 81.61% -> 78.32% | Hurt |
| Account for privacy cost of hyperparameter tuning (honest vs dishonest DP-SGD) | DP-SGD · Synthetic-L, eps=8 | test accuracy (%): 95.43% -> 95.17 % | Hurt |
| Dishonest DP-SGD (no tuning cost) compared with Algorithm 1 | DP-SGD vs Algorithm 1 · Synthetic-L, eps=0.1 | test accuracy (%): 93.85% -> 90.05% | Hurt |
| Dishonest DP-SGD (no tuning cost) compared with Algorithm 1 | DP-SGD vs Algorithm 1 · Adult, eps=1 | test accuracy (%): 83.54% -> 83.18% | Hurt |
| Increase privacy budget epsilon from 0.1 to 8 (gap between Algorithm 1 and honest DP-SGD narrows) | Algorithm 1 vs honest DP-SGD · Synthetic-L | test accuracy (%): 90.05% -> 95.30% vs 77.34% -> 95.17 % | Mixed |
Closest papers by meaning
- An Improved Privacy and Utility Analysis of Differentially Private SGD with Bounded Domain and Smooth Losses (AAAI 2026 · similarity 0.92)
- Oracle Efficient Private Non-Convex Optimization (ICML 2020 · similarity 0.91)
- The importance of feature preprocessing for differentially private linear optimization (ICLR 2024 · similarity 0.90)
- Differentially Private Stochastic Gradient Descent with Fixed-Size Minibatches: Tighter RDP Guarantees with or without Replacement (NeurIPS 2024 · similarity 0.89)
- Certification for Differentially Private Prediction in Gradient-Based Training (ICML 2025 · similarity 0.89)
- Nearly Tight Black-Box Auditing of Differentially Private Machine Learning (NeurIPS 2024 · similarity 0.89)
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.
