CS 598: Principles of Generative AI

Spring 2024

Course description:
Recent advancements in generative AI have equipped machine learning algorithms with the ability to learn from and accurately replicate observed data, creating new, similar data instances. This course provides an in-depth exploration of the key algorithmic developments in generative models, together with their underlying mathematical principles. We will cover a range of topics such as normalizing flows, variational autoencoders, Langevin algorithms, generative adversarial networks, diffusion models, and sequence generation models, etc.
Class time:
Tue, Thu 2:00–3:15pm (1035 - Campus Instructional Facility)
Instructor:
Prof. Tong Zhang (tozhang@illinois.edu)
- Office: Siebel Center 2118
- Office hour: Tue, Thu 3:15pm-4:30pm
TA:
Jiaqi Guan (jiaqi@illinois.edu)
- Office hour: Wed 1:00pm-2:00pm. Location: Siebel 0222
Prerequisites:
knowledge of machine learning, linear algebra, calculus, and probability, python programming
Course Material:
- Lecture slides
- Papers

Lectures:

Lecture Number	Topic
1.	Introduction
2.	Basic Neural Network Models and Optimization
3.	Energy-based Model
4.	Variational Inference
5.	Encoder, Decoder, and Auto Encoder
6.	Variational Auto Encoder
7.	Normalizing Flow
8.	Variational Normalizing Flow and Sampling Basics
9.	Markov Chain Monte Carlo
10.	MCMC and Langevin Algorithms
11.	Hamiltonian Monte Carlo and Under-damped Langevin Algorithm
12.	Distance, Generation, and Convergence
13.	Score Matching
14.	Diffusion Model (basics)
15.	Diffusion Model (reverse process)
17.	Diffusion Model Practice
18.	GAN Basics
19.	GAN Practice
20.	Neural ODE
21.	Disentanglement and Representation Learning
22.	Basic Sequence Models (pre Transformer)
23.	Transformer
24.	Sequence based Image Generation
25.	Encoder Only, Encoder Decoder, and Decoder Only Sequence Models
26.	Prompt, Finetuning, and Alignment
27.	Visual Foundation Model and Visual Text Alignment
28.	Student Project Presentation (I)
29.	Student Project Presentation (II)