Research map / Federated learning and privacy / Differential privacy
Classification Under Local Differential Privacy with Model Reversal and Model Averaging
Caihong Qin, Yang Bai
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 |
|---|---|---|---|
| add model reversal to weak classifiers | Weak classifier · Diabetes (eps=0.5) | misclassification rate (%): 49.23 -> 40.23 | Improved |
| add model averaging to weak classifiers | Weak classifier · Diabetes (eps=0.1) | misclassification rate (%): 49.55 -> 46.22 | Improved |
| combine model reversal and averaging (MRMA) | Weak classifier · Diabetes (eps=0.1) | misclassification rate (%): 49.55 -> 43.44 | Improved |
| combine model reversal and averaging (MRMA) | Weak classifier · Employee (eps=0.5) | misclassification rate (%): 49.02 -> 34.43 | Improved |
| use MRMA instead of pooling all client data | All data classifier · Diabetes (eps=0.1) | misclassification rate (%): 50.82 -> 43.44 | Improved |
| use MRMA instead of voting ensemble | Voting · Employee (eps=0.1) | misclassification rate (%): 58.23 -> 44.15 | Improved |
| use MRMA instead of histogram LDP classifier | Histogram · Diabetes (eps=1000) | misclassification rate (%): 39.80 -> 23.03 | Improved |
| MRMA at large eps (weak privacy) versus pooling all data | All data classifier · Diabetes (eps=1000) | misclassification rate (%): 22.05 vs 23.03 | No clear effect |
| add max-abs or Tanh rescaling after dimension reduction for logistic | Logistic on B-spline coefficients · Simulated functional data (d=4) | misclassification rate (%): 17.81 -> 12.32 | Improved |
| increase B and N1 (Case 1 to Case 2) (ablation) | MA classifiers · Simulated functional data | misclassification rate | Improved |
| use smaller n0 (Case 2 to Case 3) (ablation) | MA classifiers · Simulated functional data | misclassification rate | Improved |
| use larger n1 (Case 4 to Case 2) (ablation) | MA classifiers · Simulated functional data | misclassification rate | Improved |
Closest papers by meaning
- Learning from End User Data with Shuffled Differential Privacy over Kernel Densities (ICLR 2025 · similarity 0.89)
- Certification for Differentially Private Prediction in Gradient-Based Training (ICML 2025 · similarity 0.89)
- Position: Considerations for Differentially Private Learning with Large-Scale Public Pretraining (ICML 2024 · similarity 0.88)
- Attack-Aware Noise Calibration for Differential Privacy (NeurIPS 2024 · similarity 0.88)
- Optimal Differentially Private Model Training with Public Data (ICML 2024 · similarity 0.88)
- The Limits of Differential Privacy in Online Learning (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.
