Research map / Federated learning and privacy / Differential privacy
Locally Private Estimation with Public Features
Yuheng Ma, Hanfang Yang, Ke Jia
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 CART split rule instead of max-edge (ablation) | HistOfTree · Real data eps=2 (aligned) | rank sum: 22 -> 15 | Improved |
| use max-edge split rule instead of CART (ablation) | HistOfTree · Real data eps=2 (personalized) | rank sum: 27 -> 20 | Improved |
| select s, p, t by the data-driven rule (ablation) | HistOfTree (ME) · Real data eps=2 (personalized) | rank sum: 20 -> 15 | Improved |
| select s, p, t by the data-driven rule (ablation) | HistOfTree (CART) · Real data eps=2 (personalized) | rank sum: 27 -> 23 | Improved |
| use public features via HistOfTree instead of fully private Hist | Hist · Real data eps=2 (aligned) | rank sum: 45 -> 15 | Improved |
| use HistOfTree instead of obfuscating features with KRR | KRR · Real data eps=2 (aligned) | rank sum: 43 -> 15 | Improved |
| use HistOfTree instead of label-noise decision tree ParDT | ParDT · Real data eps=2 (aligned) | rank sum: 28 -> 15 | Improved |
| use AdHistOfTree instead of ParDT | ParDT · Real data eps=2 (personalized) | rank sum: 48 -> 15 | Improved |
| increase number of public features s* | HistOfTree · synthetic, n=10000 | MSE | Improved |
| increase privacy budget epsilon | HistOfTree · synthetic | MSE | Improved |
| increase tree depth p (ablation) | HistOfTree · synthetic, n=10000, eps=4 | MSE | Mixed |
| set leaf-size parameter t=2 (ablation) | HistOfTree · synthetic, n=10000, eps=4, s*=2 | MSE | Improved |
Closest papers by meaning
- Public-data Assisted Private Stochastic Optimization: Power and Limitations (NeurIPS 2024 · similarity 0.88)
- Learning from End User Data with Shuffled Differential Privacy over Kernel Densities (ICLR 2025 · similarity 0.88)
- Optimal Differentially Private Model Training with Public Data (ICML 2024 · similarity 0.88)
- Certification for Differentially Private Prediction in Gradient-Based Training (ICML 2025 · similarity 0.87)
- Locally Optimal Private Sampling: Beyond the Global Minimax (NeurIPS 2025 · similarity 0.87)
- The importance of feature preprocessing for differentially private linear optimization (ICLR 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.
