In this MIT Computer Science class, students will delve into the world of flow matching and diffusion models, the cutting edge of generative AI. The course covers a broad range of data types like images, videos, and even protein structures. Students will build a toy image diffusion model from scratch and gain hands-on experience with stochastic differential equations. The course notes provide a comprehensive explanation of all material, while lecture slides offer visual aids. Labs offer practical exercises to reinforce learning. Notably, the course excludes coverage of large language models (LLMs) as it focuses on continuous data rather than discrete data like text.
https://diffusion.csail.mit.edu