Skip to content

AIC313/CS378: Introduction to Generative Models

Minhyuk Sung, KAIST, Fall 2026


Teaser1

Time & Location

Time: Mon/Wed 10:30 a.m. - 11:45 a.m. (KST)
Location: E3-5 Room 210

Description

Generative models aim to learn complex data distributions and generate new samples, providing the foundation for a wide range of applications in computer vision, computer graphics, natural language processing, and scientific discovery. With the rapid development of deep learning, diverse generative modeling approaches have emerged, each offering distinct perspectives on learning and sampling from data distributions. In this course, we will cover the fundamental principles and recent advances in generative models, including autoregressive models, variational autoencoders, generative adversarial networks, normalizing flows, diffusion and score-based models, and flow matching. We will also discuss conditional generation, inference-time guidance, practical applications, and remaining challenges in the field.

Prerequisites

  • Solid background in machine learning and deep learning
  • Hands-on experience with neural network implementation
  • Recommended prior courses:
    • MAS.20050 Probability and Statistics
    • MAS.20001 Differential Equations and Applications
    • CS.30701 Introduction to Deep Learning

Course Staff

Instructor: Minhyuk Sung (mhsung@kaist.ac.kr)

Course Assistants:

Textbook

The main reference for this course is:

Kevin P. Murphy, Probabilistic Machine Learning: Advanced Topics, MIT Press.
[Download Link]

We will mainly use selected material from Part I: Fundamentals and Part IV: Generation, supplemented with recent papers and lecture notes where appropriate.

Grading

  • Participation & Quizzes: 15%
  • Exams: 45%
  • FastGen Challenge: 20%
  • CreativeGen Challenge: 20%

Important Dates

ALL ASSIGNMENTS ARE DUE 23:59 KST.

(Subject to Change)

  • Project Team Sign-Up: Due Sep 28 (Mon)
  • FastGen Challenge Mid-Term Evaluation (Optional): Due Oct 31 (Sat)
  • FastGen Challenge Submission: Due Nov 07 (Sat)
  • CreativeGen Challenge Submission: Due Dec 05 (Sat)

Schedule

(Subject to Change)

Week Date Topic Related Sections Links
1 Aug 31 Course Introduction Ch. 20.1–20.4 Slides
Sep 02 Probability Background Ch. 2.1.2–2.1.3, 2.1.5–2.1.6;
Ch. 2.2.1.2, 2.2.2.1; Ch. 2.3.1.1;
Ch. 3.2.1.1 		Slides
Recording
2 Sep 07 Autoregressive Models Ch. 22.1–22.4 Slides
Recording
Sep 09 Variational Autoencoders 1 Ch. 21.1–21.2.2 Slides
Recording
3 Sep 14 Variational Autoencoders 2 Ch. 21.2.3–21.2.4 Slides
Recording
Sep 16 Normalizing Flows 1 Ch. 23.1–23.2.2 Slides
Recording
4 Sep 21 FastGen / KCloud Session
Sep 23 No Class (Break) —
5 Sep 28 Normalizing Flows 2 Ch. 23.2.3-23.2.4 and 23.3
Sep 30 Generative Adversarial Networks Ch. 26.1–26.3.3
6 Oct 05 No Class (Substitute Holiday for National Foundation Day) —
Oct 07 Energy-Based Models 1 Ch. 24.1–24.2
7 Oct 12 Energy-Based Models 2 Ch. 24.3–24.4
Oct 14 Midterm Wrap-Up
8 Oct 19 No Class (Midterm Week) —
Oct 21 No Class (Midterm Week) —
9 Oct 26 Diffusion Models 1 Ch. 25.1–25.2.2
Oct 28 Diffusion Models 2 Ch. 25.2.3–25.3.3
10 Nov 02 Diffusion Models 3 Ch. 25.4–25.5.3
Nov 04 Conditional Generation / Latent Diffusion Ch. 25.5.4 and 25.6
11 Nov 09 Flow Matching 1 Ch. 23.2.6 and 25.4
Nov 11 Flow Matching 2
12 Nov 16 FastGen Recap / CreativeGen
Nov 18 Guest Lecture 1: Jiaxin Shi
Meta SuperIntelligence Labs
Discrete Diffusion
13 Nov 23 Inference-Time Scaling
Nov 25 Recent Topics / Course Wrap-Up
14 Nov 30 Guest Lecture 2: Gyu Rie Lee
KAIST
Generative Protein Design
Dec 02 Guest Lecture 3: Julius Berner
NVIDIA
Fast Video Generation
15 Dec 07 Project Presentations 1
Dec 09 Project Presentations 2
16 Dec 14 No Class (Final Week) —
Dec 16 No Class (Final Week) —

AI Coding Assistant Tool Policy

You are allowed (and even encouraged) to utilize AI coding assistant tools, such as ChatGPT, Copilot, Codex, and Code Intelligence, for your programming assignments and projects. Utilizing AI coding assistant tools will not be deemed as plagiarism. However, it is still strictly prohibited to directly copy code from the Internet or from someone else. Doing so will lead to a score of zero and a report to the university.


  1. AI-generated using GPT
     ↩