Research map / Federated learning and privacy / Differential privacy
Private Learning with Public Feature Conditioning
Shuli Jiang, Walid Krichene, Nicolas Mayoraz
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 |
|---|---|---|---|
| condition public-feature embeddings with C = V Sigma^-1 V^T (Cond-DP) | DPSGD · synthetic private linear regression | validation MSE | Improved |
| increase spectral decay of public features (larger p) (ablation) | Cond-DP vs DPSGD · synthetic private linear regression | validation MSE | Improved |
| apply conditioning when singular values are uniform (p = 0) (ablation) | Cond-DP · synthetic private linear regression | validation MSE | No clear effect |
| condition public features with Cond-DP | DPSGD · real-world linear regression (wine, Boston, energy, CA housing) | validation MSE | Improved |
| switch from conditioning to DPSGD after early epochs (Switch-Cond-DP) | two-layer MLP regression · supercond, TomsHardware, wave | validation MSE | Improved |
| keep conditioning throughout training instead of switching to DPSGD (ablation) | Cond-DP on MLP models · MLP regression | loss reduction | Hurt |
| use Cond-DP instead of DPSGD | linear model · Criteo Sponsored Search Conversion | validation MSE | Improved |
| use Cond-DP instead of RR-on-Bins | linear model · Criteo Sponsored Search Conversion | validation MSE | Mixed |
| use Switch-Cond-DP instead of DPSGD/RR-on-Bins | MLP (128, 64) · Criteo Sponsored Search Conversion | validation MSE | Mixed |
Closest papers by meaning
- The importance of feature preprocessing for differentially private linear optimization (ICLR 2024 · similarity 0.89)
- Public-data Assisted Private Stochastic Optimization: Power and Limitations (NeurIPS 2024 · similarity 0.89)
- Revisiting Differentially Private ReLU Regression (NeurIPS 2024 · similarity 0.88)
- Optimal Differentially Private Model Training with Public Data (ICML 2024 · similarity 0.88)
- Enhancing Learning with Label Differential Privacy by Vector Approximation (ICLR 2025 · similarity 0.88)
- Differentially Private High-dimensional Variable Selection via Integer Programming (NeurIPS 2025 · 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.
