Research map / Federated learning and privacy / Differential privacy
Scalable Private Learning with PATE
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 |
|---|---|---|---|
| Replace Laplace vote noise with Gaussian noise | PATE aggregation on Glyph · Glyph | privacy cost at fixed label accuracy | Helped |
| Use Confident-GNMax instead of LNMax | PATE student · MNIST | accuracy: 98.0% -> 98.5% | Helped |
| Use Confident-GNMax instead of LNMax | PATE student · SVHN | accuracy: 82.7% -> 91.6% | Helped |
| Use Confident-GNMax instead of LNMax | PATE student · Adult | accuracy: 83.0% -> 83.7% | Helped |
| Use Confident-GNMax instead of LNMax | PATE student · Glyph | accuracy: 72.4% -> 75.5% | Helped |
| Use Confident-GNMax instead of LNMax | PATE student · MNIST | privacy bound ε: 2.04 -> 1.97 | Helped |
| Use Confident-GNMax instead of LNMax | PATE student · SVHN | privacy bound ε: 5.04 -> 4.96 | Helped |
| Use Confident-GNMax instead of LNMax | PATE student · Adult | privacy bound ε: 2.66 -> 1.90 | Helped |
| Use Confident-GNMax instead of LNMax | PATE student · Glyph | privacy bound ε: 4.3 -> 2.03 | Helped |
| Increase the number of teachers | GNMax ensemble · Glyph | privacy-utility tradeoff | Helped |
| Use Interactive-GNMax in a second training round | Glyph student · Glyph | privacy bound ε: 0.837 | Helped |
| Raise the noisy consensus threshold from moderate to aggressive (ablation) | Confident-GNMax query selection · Glyph | number of answered queries | Mixed |
Closest papers by meaning
- Hot PATE: Private Aggregation of Distributions for Diverse Tasks (ICLR 2026 · similarity 0.94)
- Certification for Differentially Private Prediction in Gradient-Based Training (ICML 2025 · similarity 0.86)
- Learning from End User Data with Shuffled Differential Privacy over Kernel Densities (ICLR 2025 · similarity 0.86)
- It's My Data Too: Private ML for Datasets with Multi-User Training Examples (ICML 2025 · similarity 0.86)
- Position: Considerations for Differentially Private Learning with Large-Scale Public Pretraining (ICML 2024 · similarity 0.85)
- Improved Bounds for Pure Private Agnostic Learning: Item-Level and User-Level Privacy (ICML 2024 · similarity 0.85)
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.