Research map / Federated learning and privacy / Differential privacy
Efficient, Secure, Differentially Private Deep Learning in the Two-Server Model
Jun Feng, Hong Sun, Pengfei Zhang, Bocheng Ren, Shunli Zhang
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 |
|---|---|---|---|
| replace ReLU with bounded clipped activation (0 to 1) | CDPDL-style DP training · MNIST (rho=8) | test accuracy (%): 96.94 -> 97.25 | Improved |
| replace ReLU with bounded clipped activation | CDPDL-style DP training · CIFAR10 (rho=17.5) | test accuracy (%): 57.25 -> 58.63 | Improved |
| replace ReLU with bounded clipped activation | CDPDL-style DP training · FashionMNIST (rho=2) | test accuracy (%): 81.83 -> 83.02 | Improved |
| use garbled-circuits-free secure activation protocol instead of ABY2.0 | secure ReLU · runtime, batch 16 | total time (s): 1013.06 -> 88.6 | Improved |
| use garbled-circuits-free secure activation protocol instead of ABY2.0 | new activation · runtime, batch 16 | total time (s): 1859.15 -> 174.21 | Improved |
| use CryptDP secure gradient clipping and other protocols instead of ABY2.0-based | end-to-end training · runtime | total time (s): 173960.9 -> 642.78 | Improved |
| use CryptDP instead of OUE/SUE local DP at much smaller privacy budget | CryptDP · FashionMNIST | test accuracy (%): 58.45 -> 86.11 | Improved |
| increase privacy loss rho from 2 to 10 (ablation) | CryptDP · MNIST | test accuracy (%): 96.52 -> 97.42 | Improved |
| use CryptDP instead of OME-based LDPDL | CryptDP · MNIST (rho=9) | test accuracy (%): 95.26 -> 97.34 | Improved |
| use CryptDP softmax protocol instead of ABY2.0 | secure Softmax · runtime, batch 16 | total time (s): 357.32 -> 175.1 | Improved |
Closest papers by meaning
- Delving into Differentially Private Transformer (ICML 2024 · similarity 0.86)
- United We Stand: Accelerating Privacy-Preserving Neural Inference by Conjunctive Optimization with Interleaved Nexus (AAAI 2024 · similarity 0.84)
- Differentially Private Prototypes for Imbalanced Transfer Learning (AAAI 2025 · similarity 0.84)
- DMM: Distributed Matrix Mechanism for Differentially-Private Federated Learning Based on Constant-Overhead Linear Secret Resharing (ICML 2025 · similarity 0.84)
- Noise-Aware Differentially Private Regression via Meta-Learning (NeurIPS 2024 · similarity 0.84)
- Rethinking DP-SGD in Discrete Domain: Exploring Logistic Distribution in the Realm of signSGD (ICML 2024 · similarity 0.84)
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.
