CSC594: Deep Generative Models
Personnel
- Instructor: Tianxiang Gao
- Meeting time: Mondays 5:45PM - 9:00PM
- Location: Lewis Center Room 1511
- Office Hours: Mondays 9:00AM-11:00AM | Zoom
- Overview: Syllabus
- Discussion: Discord
Course Description
This course explores advanced generative models in deep learning, focusing on techniques such as Autoregressive model Models, Variational Autoencoders (VAEs), Generative Adversarial Networks (GANs), Score-Based Models, and Diffusion Models. Students will learn theoretical foundations and practical applications in text generation, image synthesis, multimodal large models, and evaluating generative models. The course emphasizes hands-on assignments where students will implement and analyze models using popular frameworks like PyTorch.
Prerequisites
- DSC 578 introduces the fundamental concepts of deep learning.
- You will implement and train deep neural networks using
PyTorch, so basicPythonproficiency is required.
Textbook
No textbook is required. Materials will be drawn from classical books and recent papers. Recommended readings:
- Deep Learning book by Goodfellow, Bengio, and Courville
- Generative Deep Learning by David Foster
A list of key papers in deep learning will also be provided.
Grading
- Quizzes: 20%
- Programming Assignments: 30%
- Midterm: 20%
- Final Project: 30% (Proposal: 10%, Progress Report: 10%, Final Report: 10%)
Only the best 5 out of 10 quizzes and assignments will count toward the final grade.
Schedule
| Week | Topic | Slides | Video |
|---|---|---|---|
| 1 | Introduction | Slides | Video |
| 2 | Autoregressive Models | Slides | Video |
| 3 | Maximum Likelihood Learning | Slides | Video |
| 4 | Evaluation of Generative Models | Slides | Video |
| 5 | Variational Autoencoders (VAEs) | Slides | Video |
| 6 | Generative Adversarial Networks (GANs) | Slides | Video |
| 7 | Energy-Based Models (EBMs) | Slides | Video |
| 8 | Score-Based and Diffusion Models | Slides | Video |
| 9 | Conditional Models and Latent Space Models | Slides | Video |
| 10 | Consistency Models and Flow Matching | Slides | Video |
Additional Reading and Resources
- Review of Linear Algebra, Zico Kolter and Chuong Do, Stanford
- Review of Probability Theory, Arian Maleki and Tom Do, Stanford
- 6.S978 Deep Generative Models, Kaiming He, Fall 2024, MIT
- CS236 Deep Generative Models, Stefano Ermon, Fall 2023, Stanford
- CS294-158 Deep Unsupervised Learning, Pieter Abbeel, Spring 2024, UC Berkeley