Diffusion Models and Score-Based Generative Models Explained - From Scratch

Dive into a comprehensive 38-minute video lecture on Diffusion Models, exploring them through the lens of Score-Based Generative Models. Gain a deeper intuition for diffusion models, visualize key concepts, and understand the connections between different approaches like DDPM, DDIM, and EDM. Follow along as the lecture covers essential topics including score, score matching, noise perturbation, denoising score matching, sampling techniques, multiple noise perturbations, and differential equations. Discover the link between score-based models and diffusion models, and conclude with a summary of key insights. Enhance your understanding with references to further reading materials on sliced score matching, improved techniques for score-based generative models, and related mathematical concepts.

Introduction
Score
Score Matching
Noise Perturbation
Denoising Score Matching
Sampling
Multiple Noise Perturbations
Differential Equations
Link to diffusion models
Summary
Conclusion