New method generates differentially private synthetic data for causal inference

Researchers have developed a new method for generating differentially private synthetic data specifically tailored for causal inference. This approach, termed "causal workloads," focuses on preserving the orthogonal moments crucial for doubly robust causal estimators, unlike generic methods that prioritize overall distributional fidelity. The proposed technique can be used directly or reconstructed via maximum-entropy calibration, with a theoretical framework that decomposes ATE error into various components. Additionally, the work introduces Causal-AIM for adaptive workload selection and NA+MI for confidence intervals, enabling a single DP synthetic table to support multiple causal analyses without additional privacy costs. AI

IMPACT This research could enable more robust and privacy-preserving causal inference in AI systems by ensuring that synthetic data accurately reflects causal relationships.

RANK_REASON The cluster contains a research paper detailing a new method for differentially private synthetic data generation for causal inference.

Read on arXiv cs.LG →

• alphaXiv
• arXiv
• AT&T
• average treatment effect
• Causal-AIM
• DagsHub
• doubly robust causal estimators
• Hugging Face
• IArxiv
• maximum-entropy calibration
• NA+MI

• paper
• safety

AI-generated summary · Google Gemini · from 2 sources. How we write summaries →