Research map / Federated learning and privacy / Differential privacy
Differentially Private Bagging: Improved utility and cheaper privacy than subsample-and-aggregate
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 |
|---|---|---|---|
| partition data k=100 times (bagging) instead of once, with improved privacy accounting (DPBag vs SAA) | subsample-and-aggregate student · Heart Failure, eps=1, n=250 | accuracy: .5798 -> .5986 | Helped |
| DPBag vs SAA at eps=3 | subsample-and-aggregate · Heart Failure, n=250 | number of labels answered: 2108 -> 4044 | Helped |
| add improved data-dependent privacy bound (personalised moments accountant) to bagging (ablation) | DPBag- · Heart Failure, eps=1, n=250, k=100 | AUROC: .5884 -> .6096 | Helped |
| add bagging (multiple partitions) without improved privacy bound (DPBag- vs SAA) (ablation) | SAA · Heart Failure, eps=1, n=250, k=100 | accuracy: .5798 -> .5875 | Helped |
| increase partitions k from 50 to 100 (ablation) | DPBag · Heart Failure, n=250, eps=1 | accuracy: .5888 -> .5986 | Helped |
| increase partitions k from 50 to 100 (ablation) | DPBag · Heart Failure, n=250, eps=5 | accuracy: .6099 -> .6091 | Hurt |
| increase partitions k from 50 to 100 (ablation) | DPBag · Heart Failure, n=250, eps=5 | number of labels: 6380 -> 6805 | Helped |
| increase number of teachers n from 100 to 250 at small eps (ablation) | DPBag · Heart Failure, k=100, eps=1 | accuracy: .5667 -> .5986 | Helped |
| increase number of teachers n from 100 to 250 at large eps (ablation) | DPBag · Heart Failure, k=100, eps=5 | accuracy: .6142 -> .6091 | Hurt |
Closest papers by meaning
- Learning from End User Data with Shuffled Differential Privacy over Kernel Densities (ICLR 2025 · similarity 0.88)
- Instance-Specific Asymmetric Sensitivity in Differential Privacy (NeurIPS 2024 · similarity 0.88)
- A Generalized Shuffle Framework for Privacy Amplification: Strengthening Privacy Guarantees and Enhancing Utility (AAAI 2024 · similarity 0.88)
- On the Price of Differential Privacy for Hierarchical Clustering (ICLR 2025 · similarity 0.88)
- Differential Privacy Under Class Imbalance: Methods and Empirical Insights (ICML 2025 · similarity 0.87)
- Certification for Differentially Private Prediction in Gradient-Based Training (ICML 2025 · 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.