Research map / Federated learning and privacy / Differential privacy
Data-dependent PAC-Bayes priors via differential privacy
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 a private data-dependent Gaussian prior instead of the data-independent prior in the Lever PAC-Bayes bound | Two-stage SGLD on MNIST · MNIST test | PAC-Bayes bound: 0.21 vs 0.26 | Helped |
| Use a private data-dependent Gaussian prior instead of the data-independent prior in the Lever PAC-Bayes bound | Two-stage SGLD on MNIST · MNIST test | PAC-Bayes bound: 0.35 vs 1 | Helped |
| Use a private data-dependent Gaussian prior instead of the data-independent prior in the Lever PAC-Bayes bound | Two-stage SGLD on MNIST · MNIST test | PAC-Bayes bound: 0.65 vs 1 | Helped |
| Increase the second-stage inverse temperature to 3*10^5 | Two-stage SGLD with DP PAC-Bayes bound · MNIST test | PAC-Bayes bound: 1 vs 1 | No clear effect |
| Increase the second-stage inverse temperature from 3*10^3 to 3*10^5 | Two-stage SGLD · MNIST test | 0-1 error: 0.12 -> 4 | Hurt |
Closest papers by meaning
- PAC-Bayes Learning Bounds for Sample-Dependent Priors (NeurIPS 2020 · similarity 0.92)
- On Inherent Privacy of Posterior Sampling: A Unified R\'enyi-Divergence Framework (NeurIPS 2026 · similarity 0.88)
- PAC-Bayes-Chernoff bounds for unbounded losses (NeurIPS 2024 · similarity 0.88)
- Privacy Guarantees in Posterior Sampling under Contamination (NeurIPS 2026 · similarity 0.87)
- Computationally Efficient Replicable Learning of Parities and Applications (NeurIPS 2026 · similarity 0.87)
- Recursive PAC-Bayes: A Frequentist Approach to Sequential Prior Updates with No Information Loss (NeurIPS 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.