Diffusion models have established themselves as state-of-the-art in generative modeling and likelihood estimation of high-dimensional image data. We explain their theoretical foundations, derive the variational lower bound, and draw connections to variational auto-encoders, normalizing flows, and stochastic optimal control. The underlying concepts are largely based on works by Song et al. (2020), Kingma et al. (2021), and Huang et al. (2021).

Resources:

• Song 2020
• Kingma 2021
• Huang 2021