Research map / Federated learning and privacy / Differential privacy
ACTG-ARL: Differentially Private Conditional Text Generation with RL-Boosted Control
Yuzheng Hu, Ryan McKenna, Da Yu, Shanshan Wu, Han Zhao, Zheng Xu, Peter Kairouz
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 AIM tabular synthesizer instead of DP-FT as feature generator (ablation) | ACTG Stage 1 · bioRxiv | feature learning JS distance | Improved |
| use DP-FT conditional generator instead of direct prompting (ablation) | ACTG Stage 2 · bioRxiv | conditional generation JS distance | Improved |
| use rich tabular schema S3 instead of free-form summary S2 (ablation) | ACTG · bioRxiv, PMC-patients | MAUVE and utility | Improved |
| use 8-field schema S3 instead of 3-field simple schema S3' (ablation) | ACTG (DP-FT feature generator) · bioRxiv | MAUVE and feature distribution matching | Improved |
| add RL (PPO) to ACTG conditional generator | ACTG · bioRxiv | MAUVE | Hurt |
| add SFT anchor on best-of-N data to RL (Anchored RL) (ablation) | ACTG-RL · bioRxiv | IFAcc and MAUVE | Improved |
| replace Gemini oracle with Qwen2.5-32B-Instruct (ablation) | ACTG · bioRxiv | MAUVE | No clear effect |
| use two-stage instead of single-stage conditional generation (ablation) | ACTG framework (S2) · bioRxiv | MAUVE and attribute matching | Improved |
| use weaker embedding models for MAUVE (ablation) | MAUVE evaluation · bioRxiv | MAUVE | Mixed |
Closest papers by meaning
- SynBench: A Benchmark for Differentially Private Text Generation (NeurIPS 2026 · similarity 0.89)
- Synthesizing Privacy-Preserving Text Data via Finetuning *without* Finetuning Billion-Scale LLMs (ICML 2025 · similarity 0.88)
- Struct-Bench: A Benchmark for Differentially Private Structured Text Generation (NeurIPS 2025 · similarity 0.88)
- Data-adaptive Differentially Private Prompt Synthesis for In-Context Learning (ICLR 2025 · similarity 0.88)
- PrE-Text: Training Language Models on Private Federated Data in the Age of LLMs (ICML 2024 · similarity 0.87)
- Masked Language Modeling Becomes Conditional Density Estimation for Tabular Data Synthesis (AAAI 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.
