Research map / Federated learning and privacy / Privacy attacks and secure learning
Privasis: Synthesizing the Largest "Public" Private Dataset from Scratch
Hyunwoo Kim, Niloofar Mireshghallah, Michael Duan, Rui Xin, Shuyue Stella Li, Jaehun Jung, David Acuna, Qi Pang, Hanshen Xiao, G. Edward Suh, Sewoong Oh, Yulia Tsvetkov and 2 more
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 |
|---|---|---|---|
| train on Privasis-Sanitization (37K subset) | Qwen3-4B · Privasis-Sanitization Vanilla test | Full Successful Record (%): 53.65 -> 72.50 | Improved |
| train on Privasis-Sanitization (37K subset) | Qwen3-0.6B · Privasis-Sanitization Vanilla test | Full Successful Record (%): 16.70 -> 68.04 | Improved |
| train on Privasis-Sanitization (37K subset) | Qwen3-4B · Privasis-Sanitization Hard test | Full Successful Record (%): 5.66 -> 12.36 | Improved |
| train on Privasis-Sanitization (37K subset) | Qwen3-0.6B · Privasis-Sanitization Hard test | Full Successful Record (%): 4.44 -> 9.31 | Improved |
| train on Privasis-Sanitization (37K subset) | Qwen3-4B · Privasis-Sanitization Vanilla test | Retention Successful Record (%): 90.02 -> 98.66 | Improved |
| train on Privasis instead of NaP2 | Qwen3-4B · Privasis-Sanitization Vanilla test | Full Successful Record (%): 32.0 -> 72.5 | Improved |
| train on Privasis instead of NaP2 | Qwen3-4B · NaP2 (zero-shot) | leak ratio (%): 10.0 -> 10.0 | No clear effect |
| add record type and format control variables (ablation) | Privasis synthesis · Privasis synthesis ablation (500 records, GPT-OSS-120B) | cosine similarity: 0.408 -> 0.322 | Improved |
| add record type and format control variables (ablation) | Privasis synthesis · Privasis synthesis ablation (500 records, GPT-OSS-120B) | Vendi score: 39.4 -> 57.3 | Improved |
| add iterative revision (ablation) | Privasis synthesis · Privasis synthesis ablation (500 records, GPT-OSS-120B) | Vendi score: 53.2 -> 57.3 | Improved |
| add diversity-preserving term to revision (ablation) | Privasis synthesis · Privasis synthesis ablation (500 records, GPT-OSS-120B) | Vendi score: 50.1 -> 57.3 | Improved |
| add diversity-preserving term to revision (ablation) | Privasis synthesis · Privasis synthesis ablation (500 records, GPT-OSS-120B) | cosine similarity: 0.354 -> 0.322 | Improved |
Closest papers by meaning
- Synthesizing Privacy-Preserving Text Data via Finetuning *without* Finetuning Billion-Scale LLMs (ICML 2025 · similarity 0.87)
- PrE-Text: Training Language Models on Private Federated Data in the Age of LLMs (ICML 2024 · similarity 0.87)
- SynBench: A Benchmark for Differentially Private Text Generation (NeurIPS 2026 · similarity 0.86)
- A Synthetic Dataset for Personal Attribute Inference (NeurIPS 2024 · similarity 0.86)
- Beyond Memorization: Violating Privacy via Inference with Large Language Models (ICLR 2024 · similarity 0.85)
- PrivAuditor: Benchmarking Data Protection Vulnerabilities in LLM Adaptation Techniques (NeurIPS 2024 · similarity 0.85)
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.
