Research map / Federated learning and privacy / Differential privacy
Improved Differential Privacy for SGD via Optimal Private Linear Operators on Adaptive Streams
Serguei Denissov, Hugh Brendan McMahan, J Keith Rush, Adam Smith, Abhradeep Guha Thakurta
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 |
|---|---|---|---|
| iterate the fixed-point mapping instead of prior algorithm or gradient descent | optimal prefix-sum factorization · 2048x2048 prefix-sum matrix | time to reach loss (minutes): over 80 minutes -> less than 3 minutes | Improved |
| learning rate decay for the last 512 rounds (0.15x) | matrix factorization mechanisms · StackOverflow next-word prediction | test accuracy | Improved |
| linear learning rate cooldown from 1.0x to 0.05x over last 512 rounds | matrix factorization mechanisms · StackOverflow next-word prediction | test accuracy | Improved |
| increase clients per round from 100 to 167 | cooldown mechanism · StackOverflow next-word prediction | test accuracy: 100 -> 167 clients/round | Improved |
| matrix factorization mechanisms with constant learning rate vs previous state of the art | DP-FTRL · StackOverflow next-word prediction | test accuracy | Improved |
Closest papers by meaning
- Multi-Epoch Matrix Factorization Mechanisms for Private Machine Learning (ICML 2023 · similarity 0.89)
- Instance-optimality in differential privacy via approximate inverse sensitivity mechanisms (NeurIPS 2020 · similarity 0.89)
- Adaptive Methods Are Preferable in High Privacy Settings: An SDE Perspective (ICLR 2026 · similarity 0.89)
- DMM: Distributed Matrix Mechanism for Differentially-Private Federated Learning Based on Constant-Overhead Linear Secret Resharing (ICML 2025 · similarity 0.87)
- Sampling-Free Privacy Accounting for Matrix Mechanisms under Random Allocation (NeurIPS 2026 · similarity 0.87)
- Improved Communication-Privacy Trade-offs in $L_2$ Mean Estimation under Streaming Differential Privacy (ICML 2024 · similarity 0.87)
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.
