It does what it says on the tin: safe synthetic data from coarsened margins
This paper introduces a novel method for generating synthetic data that offers enhanced transparency and data privacy. The approach ensures users understand which original data relationships are preserved in the synthetic version. It achieves this by first applying statistical disclosure control to relevant data margins and then using these adjusted margins to create synthetic data via the Iterative Proportional Fitting algorithm. AI
IMPACT Provides a new technique for generating synthetic data, potentially improving privacy and utility in AI model training.