ECE 598ZZ: Generative Modeling with Diffusion and Flow Matching Models

Course Overview

Generative modeling is a critical area in machine learning that focuses on learning data distributions and generating new, realistic data samples. In recent years, diffusion models and flow matching models have emerged as powerful approaches for building high-quality generative models, demonstrating success in image generation, text-to-image synthesis, and various other domains.

This course provides a deep dive into the theory, methodology, and practical applications of diffusion models and flow matching models, as well as how they compare to other generative modeling techniques. Students will learn to implement and experiment with these models, understand their underlying principles, and apply them to real-world tasks such as image generation, superresolution, and generating protein ensembles. The students will also explore intriguing open problems within the field of generative modeling and be encouraged to develop innovative solutions and new perspectives.

• Lecture Times: Tuesdays & Thursdays, 9:30 AM - 10:50 AM
• Instructor: Zhizhen Zhao
• Location: 2017 ECEB
• Office Hour: Tue 11:00 AM - 12:00 PM | Location: CSL 109
• TA: Yichi Zhang | Office Hour: TBD | Location: TBD
• Recorded lectures are available at media space
• Campuswire and gradescope entry codes
• Chicago campus students: Zoom info

Schedule

Assignments

• Homework 1
• Homework 2
• Literature review (latex template).  4 page limit for the main paper (not including references and appendices).
• Project (latex template).  4 page limit for the main paper (not including references and appendices).
• Presentation. 10 minutes total team presentation. (Schedule TBD). 

Resources

• Syllabus
• Reading list