Diffusion Models and Flow Matching
Guest lecture, George Washington University, 2026
I gave a guest lecture on diffusion models and flow matching for MAE 6291 — Large Language and Vision Models at George Washington University on September 9, 2026 (Fall 2026).
Related writing: Generative AI: Diffusion & Flow Models from First Principles.
From Noise to Data
The lecture introduces generation as sampling, then builds flow models through velocity fields and numerical integration. It covers conditional flow matching, the connection to diffusion and score functions, and classifier-free guidance.
By the end, students should be able to explain how a generator can be trained using regression and sampled using numerical integration, and how diffusion fits into that framework.
View the lecture slides (PDF).
The slides credit Peter Holderrieth and Ezra Erives’ MIT 6.S184 course for the flow-first exposition and selected illustrations, with additional examples and annotations.