Authors present Anonymeter, a statistical framework to jointly quantify different types of privacy risks in synthetic tabular datasets. We equip […]
The results show that diffusion models are much less private than prior generative models and that mitigating these vulnerabilities may require new advances in privacy-preserving training.