Research map / Federated learning and privacy / Differential privacy
Practical and Private (Deep) Learning Without Sampling or Shuffling
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 |
|---|---|---|---|
| Use DP-FTRL with correlated tree-aggregation noise instead of unamplified DP-SGD | DP training · MNIST, CIFAR-10, and EMNIST | test accuracy | Helped |
| Use DP-FTRL with correlated tree-aggregation noise instead of amplified DP-SGD at the same computation cost | DP training · MNIST, CIFAR-10, and EMNIST | test accuracy | Mixed |
| Increase DP-FTRL minibatch size fourfold (ablation) | DP-FTRLM · MNIST, CIFAR-10, and EMNIST | test accuracy | Helped |
| Remove privacy amplification (ablation) | DP-SGD · MNIST, CIFAR-10, and EMNIST | test accuracy | Hurt |
| Increase unamplified DP-SGD minibatch size fourfold (ablation) | DP-SGD · MNIST, CIFAR-10, and EMNIST | test accuracy | Hurt |
| Add momentum 0.9 (ablation) | DP-FTRL · Image classification and StackOverflow | test accuracy | Helped |
| Use DP-FTRLM instead of DP-SGDM at the same computation cost | StackOverflow user-level DP training · StackOverflow | test accuracy | Mixed |
| Increase DP-FTRLM report goal fourfold (ablation) | StackOverflow user-level DP training · StackOverflow | test accuracy | Helped |
| Increase DP-SGDM report goal fourfold without privacy amplification (ablation) | DP-SGDM · StackOverflow user-level DP training | test accuracy | Helped |
| Use DP-FTRLM instead of DP-SGDM at fixed accuracy targets | StackOverflow user-level DP training · StackOverflow, 24.5% accuracy target | computational cost | Helped |
| Use DP-FTRLM instead of DP-SGDM at fixed accuracy targets | StackOverflow user-level DP training · StackOverflow, 23% accuracy target | computational cost | Helped |
| Use DP-FTRL-SometimesRestart for many training epochs (ablation) | DP-FTRL · CIFAR-10 and EMNIST | test accuracy | Mixed |
Closest papers by meaning
- Differentially Private Stochastic Gradient Descent with Fixed-Size Minibatches: Tighter RDP Guarantees with or without Replacement (NeurIPS 2024 · similarity 0.91)
- Privacy-Preserving Federated Convex Optimization: Balancing Partial-Participation and Efficiency via Noise Cancellation (ICML 2025 · similarity 0.90)
- Scalable DP-SGD: Shuffling vs. Poisson Subsampling (NeurIPS 2024 · similarity 0.90)
- Mitigating the Privacy–Utility Trade-off in Decentralized Federated Learning via f-Differential Privacy (NeurIPS 2025 · similarity 0.89)
- DOPPLER: Differentially Private Optimizers with Low-pass Filter for Privacy Noise Reduction (NeurIPS 2024 · similarity 0.89)
- Towards hyperparameter-free optimization with differential privacy (ICLR 2025 · similarity 0.89)
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.