Research map / Federated learning and privacy / Differential privacy
Particle Filter for Bayesian Inference on Privatized Data
Yu-Wei Chen, Pranav Sanghi, Jordan Awan
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 |
|---|---|---|---|
| use DP-PF instead of DP-DA-MCMC | private posterior sampler · Linear regression non-conjugate prior, eps=0.5, n=500 | seconds per ESS: 1173.600 -> 0.638 | Improved |
| use DP-PF instead of DP-DA-MCMC | private posterior sampler · Linear regression non-conjugate prior, eps=0.5, n=500 | 90% HPD width of beta_1: 3.009 -> 2.097 (Reject-ABC 2.186) | Improved |
| use DP-PF instead of DP-Reject-ABC | private posterior sampler · Linear regression non-conjugate prior, eps=1.0, n=500 | seconds per ESS: 70.577 -> 1.696 | Improved |
| use DP-PF instead of DP-Reject-ABC at large n and eps | private posterior sampler · Location-scale normal, eps=2.0, n=1000 | seconds per ESS: 5.74 -> 0.07 | Improved |
| use DP-PF instead of DP-DA-MCMC with conjugate prior | private posterior sampler · Linear regression conjugate prior, eps=0.5, n=500 | seconds per ESS: 45.503 -> 0.868 | Improved |
| use DP-PF instead of DP-Reject-ABC with conjugate prior at eps=2, n=300 | private posterior sampler · Linear regression conjugate prior, eps=2, n=300 | seconds per ESS: 17.053 -> 0.591 | Improved |
| use DP-PF instead of DP-DA-MCMC with M=2 histogram layers | private posterior sampler · Gaussian mixture model, M=2 | posterior recovery of parameters | Improved |
| exponential instead of linear epsilon_t schedule (ablation) | DP-PF · Linear regression non-conjugate prior, eps=2.0, n=500 | seconds per ESS: 7.481 -> 14.021 | Hurt |
| exponential instead of linear epsilon_t schedule (ablation) | DP-PF · Linear regression non-conjugate prior, eps=0.5, n=300 | posterior mean of beta_1: 1.973 -> 1.970 | No clear effect |
| exponential instead of linear epsilon_t schedule (ablation) | DP-PF · Linear regression conjugate prior, eps=2.0, n=500 | seconds per ESS: 11.657 -> 22.512 | Hurt |
| increase number of particles N beyond 1000 (ablation) | DP-PF · Linear regression CI coverage, eps>=0.5 | 95% CI coverage and width: coverage around 0.95 and CI width below 0.05 for N > 1000 | Improved |
| smaller privacy budget eps (ablation) | DP-PF · Linear regression CI coverage | particles needed for coverage | Hurt |
Closest papers by meaning
- Privacy Guarantees in Posterior Sampling under Contamination (NeurIPS 2026 · similarity 0.91)
- On Inherent Privacy of Posterior Sampling: A Unified R\'enyi-Divergence Framework (NeurIPS 2026 · similarity 0.90)
- Analysis of Differentially Private Synthetic Data: A Measurement Error Approach (AAAI 2024 · similarity 0.87)
- Private Statistical Estimation via Truncation (NeurIPS 2025 · similarity 0.87)
- Differentially Private Equivalence Testing for Continuous Distributions and Applications (NeurIPS 2024 · similarity 0.86)
- Improving the Variance of Differentially Private Randomized Experiments through Clustering (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.
