DEEP GENERATIVE MODELS



Class overview

Deep generative models combine the generality of probabilistic reasoning with the scalability of deep learning. This research area is at the forefront of deep learning and has given state-of-the-art results in text generation, video synthesis, molecular design, amongst many others. This course will cover recent advances in deep generative models, including variational autoencoders, generative adversarial networks, autoregressive models, and normalizing flow models. This is a Ph.D. level course with emphasis on mathematical principles as well as practical know-how. The course will be a combination of lectures, student presentations, and team projects.

Lecturer: Qi (Rose) Yu (roseyu@ucsd.edu)

Syllabus

Lectures

Class Assessment

  • 30 % homework (10% x 3)

  • 50 % project

    • 10 % proposal

    • 10 % milestone report

    • 20 % final report

    • 10 % final presentation

  • 15 % paper discussion

  • 5 % lecture scribe

Resources

FAQ

Q: What are the pre-requisites?

  • Familiarity with statistical inference and deep learning.

  • Have taken at least an equivalent of CSE 250A and CSE 253.

  • Proficiency in Python.

Q: Can masters/undergraduates take this course?

  • This a PhD level audience.

  • Exceptional masters/undergraduates are welcome.

About me

My Chinese name is Qi Yu. That is also the instructor name in the registrar's office. I publish under the name Rose Yu. You can learn more about my research at my personal website.