Diffusion and flow models are the cutting edge generative AI methods for images, videos, and many other data types. This course offers a comprehensive introduction for students and researchers seeking a deeper understanding of these models. Lectures will teach the core mathematical concepts necessary to understand diffusion models, including stochastic differential equations and the Fokker-Planck equation, and will provide a step-by-step explanation of the components of each model. Labs will accompany each lecture allowing students to gain practical, hands-on experience with the concepts learned in a guided manner. At the end of the class, students will have built a latent diffusion model from scratch – and along the way, will have gained hands-on experience with the mathematical toolbox of stochastic analysis that is useful in many other fields. This course is ideal for those who want to explore the frontiers of generative AI through a mix of theory and practice. We recommend some prior experience with probability theory and deep learning..
Course Notes
The course notes serve as the backbone of the course and provide a self-contained explanation of all material in the class. We strongly recommend using them. You can view the notes by clicking the button below:
Lecture Schedule
| Lecture | Date & Time | Room | Topics | Slides | Recording |
|---|---|---|---|---|---|
| 01 | Tue, Jan 20 — 11:00–12:30 | E25-111 | Flow and Diffusion Models | Lec 01 | Link |
| 02 | Thu, Jan 22 — 11:00–12:30 | E25-111 | Flow Matching | Lec 02 | Link |
| 03 | Fri, Jan 23 — 11:00–12:30 | E25-111 | Score Matching, Guidance | Lec 03 | Link |
| 04 | Wed, Jan 28 — 11:00–12:30 | E25-111 | Latent Spaces, Neural Network architectures | Lec 04 | Link |
| 05 | Fri, Jan 30 — 12:30-2:00 | E25-111 | Discrete Diffusion Models | Lec 05 | Link |
Labs
Grading is P/F and based on class participation and completion of labs. There will be 3 labs given as homework which culminate in a codebase training a flow matching model. The labs will guide you through building this model from scratch step-by-step. To support you in completing the labs, there will be office hours where you can ask questions and get help. The office hours schedule is as follows:| Office Hour | Date & Time | Room | Associated Lab |
|---|---|---|---|
| 01 | Wed, Jan 21 — 11:00–12:30 | 36-144 | Lab 1 |
| 02 | Fri, Jan 23 — 15:00–16:30 | 36-156 | Lab 2 |
| 03 | Tue, Jan 27 — 11:00–12:30 | 36-144 | Lab 3 |
Instructors
Prerequisites: Linear algebra, multivariate calculus, and basic probability theory. Students should be familiar with Python and have some experience with PyTorch.
Questions? Email either Peter (phold@mit.edu) or Ron (ronsh@mit.edu)!
Class of 2025: To view a previous iteration of the class, click on the 2025 tab. Yet, we will have new content and a new structure this year.