Research map / Federated learning and privacy / Differential privacy
Differentially Private Set Union
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 count-based histogram (each item +1 per user) with dependent update policy (l1-DESCENT) plus Laplace noise (POLICY LAPLACE) | DPSU histogram construction · Reddit unigrams, eps=3, delta=exp(-10) | unigrams released: about 2-4x larger output than count/weighted baselines | Helped |
| Use l2-DESCENT update policy with Gaussian noise (POLICY GAUSSIAN) instead of baselines | DPSU histogram construction · Reddit unigrams, eps=3, delta=exp(-10) | unigrams released: up to 17113 vs 8996 for best weighted Gaussian | Helped |
| Increase user contribution bound Delta0 from 1 to 50 for POLICY LAPLACE (ablation) | POLICY LAPLACE · Reddit unigrams, eps=3 | unigrams released: 4482 at Delta0=1 to 15268 at Delta0=50 | Helped |
| Increase Delta0 beyond 50 for POLICY LAPLACE (ablation) | POLICY LAPLACE · Reddit unigrams, eps=3 | unigrams released: 15268 at Delta0=50 down to 13870 at Delta0=300 | Hurt |
| Increase Delta0 from 1 to 100 for POLICY GAUSSIAN (ablation) | POLICY GAUSSIAN · Reddit unigrams, eps=3 | unigrams released: 3169 at Delta0=1 to 16954 at Delta0=100 | Helped |
| Increase Delta0 for WEIGHTED GAUSSIAN from 1 to 100 (ablation) | WEIGHTED GAUSSIAN baseline · Reddit unigrams, eps=3 | unigrams released: 3194 at Delta0=1 to 8996 at Delta0=100 | Helped |
| Increase Delta0 for WEIGHTED LAPLACE from 1 to 300 (ablation) | WEIGHTED LAPLACE baseline · Reddit unigrams, eps=3 | unigrams released: 4479 at Delta0=1 declines to 3648 at Delta0=300 | Mixed |
| Increase Delta0 for COUNT LAPLACE from 1 to 300 (ablation) | COUNT LAPLACE baseline · Reddit unigrams, eps=3 | unigrams released: 4484 at Delta0=1 to 647 at Delta0=300 | Hurt |
| Use l1 vs l2 contractive policy: POLICY GAUSSIAN outperforms POLICY LAPLACE at the best Delta0 (ablation) | DPSU update policy and noise type · Reddit unigrams, eps=3, delta=exp(-10) | unigrams released: 16954 (Gaussian, Delta0=100) vs 15268 (Laplace, Delta0=50) | Helped |
Closest papers by meaning
- Incorporating Item Frequency for Differentially Private Set Union (AAAI 2022 · similarity 0.89)
- Scalable Private Partition Selection via Adaptive Weighting (ICML 2025 · similarity 0.88)
- Differentially Private Set Representations (NeurIPS 2024 · similarity 0.88)
- Differentially Private n-gram Extraction (NeurIPS 2021 · similarity 0.88)
- Differentially Private Space-Efficient Algorithms for Counting Distinct Elements in the Turnstile Model (ICML 2025 · similarity 0.85)
- Optimality of Matrix Mechanism on $\ell_p^p$-metric (ICLR 2025 · similarity 0.84)
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.