Research map / Federated learning and privacy / Privacy attacks and secure learning
Semi-supervised Knowledge Transfer for Deep Learning from Private Training Data
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 |
|---|---|---|---|
| Transfer noisy teacher labels to a student using semi-supervised GAN training | PATE · MNIST | accuracy: 93.18% -> 98.00% | Helped |
| Transfer noisy teacher labels to a student using semi-supervised GAN training | PATE · SVHN | accuracy: 87.79% -> 90.66% | Helped |
| Use PATE-G instead of the previous differentially private method | MNIST · MNIST | epsilon: 8 -> 2.04 | Helped |
| Train with distillation probabilities rather than only the plurality label (ablation) | SVHN student trained from 50 teachers · SVHN | accuracy | Helped |
| Use siamese networks to identify similar labeled and unlabeled inputs (ablation) | Active learning for the student · SVHN | teacher-query reduction | Hurt |
| Train siamese networks to match images from the same class (ablation) | Active learning for the student · SVHN | teacher-query reduction | Hurt |
| Select uncertain points using binary experts before requesting teacher labels (ablation) | Active learning for the student · SVHN | accuracy: +1.5% to +2.5% | Helped |
| Query teacher labels for the student's least-confident inputs first (ablation) | Active learning for the student · SVHN | accuracy: 4%-10% | Helped |
| Use GAN-based semi-supervised training rather than fully supervised training on a subset of labeled examples (ablation) | PATE student · MNIST and SVHN | teacher-query efficiency | Helped |
| Increase the number of teachers in the ensemble (ablation) | PATE noisy aggregation · MNIST and SVHN | accuracy | Mixed |
Closest papers by meaning
- Scalable Private Learning with PATE (ICLR 2018 · similarity 0.91)
- Certification for Differentially Private Prediction in Gradient-Based Training (ICML 2025 · similarity 0.87)
- Does Training with Synthetic Data Truly Protect Privacy? (ICLR 2025 · similarity 0.86)
- Position: Considerations for Differentially Private Learning with Large-Scale Public Pretraining (ICML 2024 · similarity 0.85)
- Differentially Private Prototypes for Imbalanced Transfer Learning (AAAI 2025 · similarity 0.85)
- Differentially Private Sum-Product Networks (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.