This repository contains lecture notes, seminar problems, homework assignments, and supplementary materials for the Diffusion-Based Generative Models course taught to Bachelor’s and Master’s students at the Faculty of Computer Science of HSE University. The course materials are also available on the FCS Wiki, which also contains organizational information about the course.
This course aims to introduce students to the foundations of diffusion models and modern generative modeling more broadly, with an emphasis on detailed mathematical derivations and applications to research-oriented problems. The materials cover, among other topics:
- stochastic differential equations, continuity and Fokker–Planck equations;
- continuous-time diffusion models, the score identity, denoising score matching, and classifier/classifier-free guidance;
- distillation of diffusion models into few-step generators, including Consistency Distillation and Distribution Matching Distillation;
- ODE solvers for efficient sampling from diffusion models;
- Flow Matching, Bridge Matching, Rectified Flow, and their connections to optimal transport;
- Schrödinger bridges as a unifying perspective on unpaired translation, sampling, and reward alignment.
Lectures: Denis Rakitin
Seminars: Alexander Oganov
Teaching assistant: Alexander Zaytsev
Current version of the lecture notes can be found at main.pdf.
Seminars and materials: coming soon.
Coming soon.