Diffusion Model의 시초인 Diffusion Probabilistic Models부터 Score-based Generative Model(NCSN), Denoising Diffusion Probabilistic Models(DDPM) 그리고 Denoising Diffusion Implicit Models(DDIM)까지 정리하는 시리즈의 첫 번째 글에서는 Diffusion Models를 위한 preliminaries와 Diffusion Probabilistic Models에 관해 리뷰한다.
1. Preliminaries
(1) Generative Model vs. Discriminative Model
(2) Explicit Density Approach vs. Implicit Density Approach in Generative Models
(3) Difficulty in estimating posterior
(4) Latent Variable Models
1) Variational Inference in Latent Variable Models
2) Variational Inference and Log Likelihood
(5) Key Papers in Diffusion Models
2. Diffusion Probabilistic Models
(1) Contribution
(2) Summary
(3) Forward Trajectory
(4) Backward Trajectory
(5) Model Probability
(6) Training
(7) Inference
(8) Key Point
(9) Problem
출처
1. Jascha Sohl-Dickstein et al., “Deep Unsupervised Learning using Nonequilibrium Thermodynamics.” arXiv:1503.03585 (2015)