Reproduce “Exploring The Empirical Privacy Landscape of Synthetic Data” on OpenResearch

Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.

Universal app for Apple silicon and Intel. Requires macOS 11 or later.

About this paper

Zihao Zhao, Avi Schwarzschild, Casey Meehan, Katherine Lee, Zhili Feng, Dylan Sam

The authors test privacy-aware rewriting, which replaces rare clinical details with plausible alternatives before model finetuning while retaining common information. In a controlled corpus of 100 synthetic canaries, GPT-5.5 rewriting reduced mean Secret-Sharer exposure versus ordinary rewriting to 2.21 bits and 1.71 bits across two target models. DP-SGD had lower exposure, and rewriting provides no formal privacy guarantee.

Summary by alphaXiv.

Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.