Research map / Federated learning and privacy / Differential privacy
Differentially Private n-gram Extraction
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 DPNE instead of DPSU-all | DPSU · Reddit | total extracted n-grams: 124,637 -> 480,191 | Helped |
| use DPNE instead of DPSU-even | DPSU · Reddit | total extracted n-grams: 109,958 -> 480,191 | Helped |
| use DPNE (budget spread over all lengths) instead of DPSU-single for 8-grams | DPSU-single · Reddit 8-grams | extracted 8-grams: 329 -> 9,019 | Helped |
| use DPNE instead of DPSU-single for 1-grams | DPSU-single · Reddit 1-grams | extracted 1-grams: 44,635 -> 16,173 | Hurt |
| use DPNE instead of DPSU-all | DPSU · MSNBC | total extracted n-grams: 4,116 -> 7,739 | Helped |
| use DPNE instead of DPSU-single on MSNBC 3-grams | DPSU-single · MSNBC 3-grams | extracted 3-grams: 1,483 -> 1,273 | Hurt |
| increase user contribution bound Delta_0 up to about 300 (ablation) | DPNE · Reddit | number of extracted n-grams | Helped |
| increase fraction of tolerated spurious n-grams eta (ablation) | DPNE · Reddit | number of extracted n-grams | Helped |
| decay noise geometrically (sigma_k = c sigma_{k-1}, c<1) instead of equal budget (ablation) | DPNE · Reddit | extracted n-grams by length | Mixed |
| use both-side instead of single-side pruning rule (ablation) | DPNE · Reddit | total extracted n-grams | No clear effect |
| use both-side instead of single-side pruning rule (ablation) | DPNE · Reddit | length distribution of n-grams | Mixed |
| compare against K-anonymity with K=100 for length 6,7,8 n-grams | DPNE · Reddit | fraction of frequent n-grams recovered: 90% | Helped |
Closest papers by meaning
- MERGE: Fast Private Text Generation (AAAI 2024 · similarity 0.85)
- SynBench: A Benchmark for Differentially Private Text Generation (NeurIPS 2026 · similarity 0.85)
- Synthesizing Privacy-Preserving Text Data via Finetuning *without* Finetuning Billion-Scale LLMs (ICML 2025 · similarity 0.85)
- InvisibleInk: High-Utility and Low-Cost Text Generation with Differential Privacy (NeurIPS 2025 · similarity 0.84)
- Scalable Private Partition Selection via Adaptive Weighting (ICML 2025 · similarity 0.84)
- Cape: Context-Aware Prompt Perturbation Mechanism with Differential Privacy (ICML 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.