Lecture 3 - Introduction to Diffusion Models (DDPM) | Principles of Diffusion Models

In this lecture, we look at the theory proposed in the paper, "Denoising Diffusion Probabilistic Models": https://arxiv.org/abs/2006.11239 We understand the diffusion process completely from scratch and view at this from a lens such that it is an extension of variational autoencoders. We first cover the forward process. Then we cover the reverse process. Here are the Google Colab notebooks which are used: Forward Transition Kernel: https://colab.research.google.com/drive/1xb2QF9j5RuLQjvCx0ap2bb9jLjr9_zsk?usp=sharing Reverse Transition Kernel: https://colab.research.google.com/drive/1J3AQEgxmsryrDQ6YIDja6oeCAkh5KtA3?usp=sharing