Efficient Provably Private Classification with a Tabular Foundation Model
PrivTab is a differentially private tabular foundation model that classifies sensitive data in one forward pass.
PrivTab embeds a privacy mechanism in a tabular foundation model pretrained on simulated data. Using in-context learning, it turns sensitive rows into compact, provably private summaries in a single forward pass. The authors report better utility than private linear and neural baselines under moderate-to-strong privacy, negligible membership leakage, calibrated predictions, and about 10,000 times faster dataset fitting.
- Existing tabular foundation models lack formal privacy and leak membership.
- PrivTab is pretrained on simulated data and learns privately in context.
- It beats private linear and neural baselines under strong privacy.
- Fitting needs one forward pass, about 10,000 times faster.
- AI models
- PrivTab
From the original article
1 min read at arxiv.orgTabular data underpin prediction and decision-making in medicine, finance, government and science, but often contain sensitive individual-level information, creating a need for accurate prediction while preserving privacy.
By Talal Alrawajfeh · arxiv.org
A short excerpt; the article and its images belong to the publisher. Summary, key facts and assessment above are AI-written from the article and can be wrong.