2020 (modified: 03 Nov 2022)ICML 2020Readers: Everyone
Abstract:We present three new algorithms for constructing differentially private synthetic data—a sanitized version of a sensitive dataset that approximately preserves the answers to a large collection of s...