Reproduce “Foundations of Bayesian Learning from 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
Harrison Wilde, Jack Jewson, Sebastian Vollmer, Chris Holmes
The authors examine how analysts should learn from differentially private synthetic data when its generating distribution differs from the target population. They compare conventional Bayesian updating with robust alternatives and propose selecting the amount of synthetic data using held-out predictive performance. Across Gaussian and logistic-regression experiments, β-divergence performs better in most cases, but not universally or across all tasks.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.