Welcome to Kun’s blog.
Hi, this is Xikun. I’m documenting my learning notes in this blog.
Notes on Diffusion Models
Notations $p(z|x)$: true posterior. $q_\phi(z|x)$: approximate posterior with parameters $\phi$. $x_0$: true data. $x_t, t \in [1,T]$: latent variables. $q(x_{t}|x_{t-1})$: forward process. $p(x_t|x_{t+1})$: reverse process. Understanding Diffusion Models from the Perspective of VAE This section briefly summarizes (Luo, 2022) with my understanding, particularly focusing on the math logic of the diffusion models. The original article by Calvin Luo is highly recommended! It helps a lot! Evidence Lower Bound What are likelihood-based generative models?...