Paper Review : Song And Ermon Thread

After summarizing Sohl-Dickstein’s 2015 paper previously, I plan to continue reading papers about diffusion models. In this thread, I will summarize the research of Yang Song and Stefano Ermon, who popularized Score-based generative modeling - another name for diffusion models. Most of the content is referenced from their papers, and I also referred to the blog post written by the author himself.

“Generative Modeling by Estimating Gradients of the Data Distribution”, NeurIPS (2019) “Score-based generative modeling through stochastic differential equation”, ICLR (2020) “Improved techniques for training score-based generative models”, NeurIPS (2020)