Extracting Training Data from Diffusion Models 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. More February 21, 2023 Publication Global Arxiv AI,research
