Research map / Federated learning and privacy / Differential privacy
Privacy Amplification via Random Check-Ins
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 no amplification with random check-ins using a fixed window | Distributed DP-SGD · One epoch DP-SGD (n updates) | central DP ε: \ensuremath{\varepsilon}_0 -> O\left( e^{1.5\ensuremath{\varepsilon}_0} / \sqrt{n}\right) | Helped |
| Use random check-ins instead of uniform subsampling | Distributed DP-SGD · One epoch DP-SGD (n updates) | central DP ε: O\left(e^{\ensuremath{\varepsilon}_0} / \sqrt{n}\right) -> O\left( e^{1.5\ensuremath{\varepsilon}_0} / \sqrt{n}\right) | Mixed |
| Use the improved shuffling analysis | Privacy amplification by shuffling · One epoch DP-SGD (n updates) | central DP ε: O\left(e^{3\ensuremath{\varepsilon}_0} / \sqrt{n}\right) -> O\left(e^{2.5\ensuremath{\varepsilon}_0} / \sqrt{n}\right) | Helped |
| Use the fixed-window random check-in bound instead of the prior shuffling bound | Privacy amplification for one epoch of DP-SGD · One epoch DP-SGD (n updates) | central DP ε: O\left(e^{3\ensuremath{\varepsilon}_0} / \sqrt{n}\right) -> O\left( e^{1.5\ensuremath{\varepsilon}_0} / \sqrt{n}\right) | Helped |
| Average updates from all checked-in clients rather than selecting one and using dummy updates | Random check-in distributed DP-SGD · Fixed-window distributed DP-SGD | central DP ε: O(\ensuremath{\varepsilon}_0 / \sqrt{m}) | Mixed |
| Restrict client check-ins to sliding availability windows | Random check-in distributed DP-SGD · Sliding-window distributed DP-SGD | central DP ε: \ensuremath{\varepsilon} \leq 7\ensuremath{\varepsilon}_0\sqrt{\frac{\log(1/\delta)}{m}} | Helped |
| Set the check-in probability to p0 = m/n and repeat the fixed-window protocol n/m times | Random check-ins with a fixed window · Repeated fixed-window protocol | central DP ε: \tilde{O}\left( e^{1.5\ensuremath{\varepsilon}_0} / \sqrt{n}\right) | Helped |
Closest papers by meaning
- Privacy-Preserving Federated Convex Optimization: Balancing Partial-Participation and Efficiency via Noise Cancellation (ICML 2025 · similarity 0.91)
- Leveraging Randomness in Model and Data Partitioning for Privacy Amplification (ICML 2025 · similarity 0.91)
- Differentially Private Stochastic Gradient Descent with Fixed-Size Minibatches: Tighter RDP Guarantees with or without Replacement (NeurIPS 2024 · 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 0.90)
- A Generalized Shuffle Framework for Privacy Amplification: Strengthening Privacy Guarantees and Enhancing Utility (AAAI 2024 · similarity 0.90)
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.