Research map / Federated learning and privacy / Differential privacy
Privacy Odometers and Filters: Pay-as-you-Go Composition
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 summed realized privacy parameters (basic composition) as a privacy odometer and filter under adaptive parameter selection | basic composition (privacy odometer/filter) · theoretical analysis (adaptive parameter composition) | validity of privacy bound | Helped |
| Naively applying advanced composition theorem to realized privacy parameters as a heuristic odometer or filter | advanced composition theorem · theoretical analysis (adaptive parameter composition) | validity of privacy bound | Hurt |
| Privacy filter based on martingale concentration (Theorem 4.1) instead of naive heuristic | advanced composition privacy filter · theoretical analysis (adaptive parameter composition) | asymptotic privacy loss bound | No clear effect |
| Privacy odometer with advanced composition bound (martingale-based) instead of filter | advanced composition privacy odometer · theoretical analysis (adaptive parameter composition) | asymptotic privacy loss bound: worse than advanced composition by at most a factor of sqrt(log log n), constant in some regimes | Mixed |
Closest papers by meaning
- Fully-Adaptive Composition in Differential Privacy (ICML 2023 · similarity 0.96)
- Adaptive Privacy Composition for Accuracy-first Mechanisms (NeurIPS 2023 · similarity 0.90)
- 𝑓-Differential Privacy Filters: Validity and Approximate Solutions (NeurIPS 2026 · similarity 0.89)
- The Last Iterate Advantage: Empirical Auditing and Principled Heuristic Analysis of Differentially Private SGD (ICLR 2025 · similarity 0.87)
- A Generalized Shuffle Framework for Privacy Amplification: Strengthening Privacy Guarantees and Enhancing Utility (AAAI 2024 · similarity 0.87)
- Auditing $f$-differential privacy in one run (ICML 2025 · similarity 0.86)
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.