Skip to content

Course Overview

Welcome to the Fall 2025 offering of the Deep Generative Models course at SUT! We are excited to have you join us on this journey into one of the most dynamic and foundational fields in modern artificial intelligence.

  • University: Sharif University of Technology (SUT)
  • Department: Department of Mathematical Sciences
  • Group: Computer Science (CS)
Course Description (Click to Expand)

This course provides a comprehensive, in-depth introduction to the principles, algorithms, and applications of deep generative modeling. We will explore a wide array of foundational and state-of-the-art architectures, beginning with core probabilistic concepts.

Key topics are structured around three main pillars:

  • Foundational Generative Models: Including Autoregressive Models, Variational Autoencoders (VAEs), Generative Adversarial Networks (GANs), Normalizing Flows, and Energy-Based Models (EBMs).
  • Modern Diffusion & Score-Based Models: A deep dive into Score-Based Models (SDEs), Flow Matching, and Diffusion Models, including their ODE/Flow formulations and applications like text-to-image generation.
  • Large Language Models (LLMs): A significant module dedicated to the principles of LLMs and LMMs, their emergent abilities, and their capacity for reasoning.
Learning Objectives (Click to Expand)

Upon successful completion of this course, students will be able to:

  • Understand the fundamental principles and theoretical underpinnings of major generative model families.
  • Implement and train deep generative models for practical applications and problem domains.
  • Master the core concepts and mathematical foundations required to analyze and compare different model architectures.
  • Develop the theoretical knowledge necessary to read, understand, and critically assess current research papers in the field.

Instructor

Tentative Schedule

Lectures

# Topic of Session Material Date
1 Introduction Topic 1
۲۰ مهر
2 Intro. to Probabilistic Graphical Models Topic 2
۲۲ مهر
3 Intro. to Probabilistic Graphical Models Topic 3
۲۷ مهر
4 Intro. to Probabilistic Graphical Models Topic 4
۲۹ مهر
5 Autoregressive Models Topic 5
۴ آبان
6 Autoregressive Models Topic 5
۶ آبان
7 VAEs Topic 6
۱۱ آبان
8 VAEs Topic 6
۱۳ آبان
9 GANs Topic 7
۱۸ آبان
10 GANs Topic 7
۲۰ آبان
11 Normalizing Flows - Invertible Models
۲۵ آبان
12 Energy Based Models
۲۷ آبان
13 Score Based Models and SDEs
۲ آذر
14 Score Based Models and SDEs
۴ آذر
15
Midterm Exam
۹ آذر
16 Flow Matching
۱۱ آذر
17 Intro. to Diffusion Models
۱۶ آذر
18 Diffusion Models and ODE/Flows
۱۸ آذر
19 Diffusion Models and ODE/Flows
۲۳ آذر
20 Text-to-Image Generation with Diffusion Models
۲۵ آذر
21 Diffusion for Discrete Data
۳۰ آذر
22 Advanced Topics in Generative Models
۲ دی
23 Intro. to LLMs and LMMs
۷ دی
24 LLM Emergent Abilities
۹ دی
25 LLM Emergent Abilities
۱۴ دی
26 Reasoning in LLMs
۱۶ دی
27 Reasoning in LLMs
۲۱ دی
28 Reasoning in LLMs
۲۳ دی

Homeworks

Homework Release Deadline
Homework 1: PGMs, ARs
۶ آبان
۲۳ آبان
Homework 2: VAEs, GANs, Flows
۲۵ آبان
۱۰ آذر
Homework 3: EBMs, SDEs, Flow Matching
۱۱ آذر
۲۹ آذر
Homework 4: Diffusion, ODEs
۳۰ آذر
۱۵ دی
Homework 5: LLMs, LMMs
۱۶ دی
۳۰ دی

Logistics & Policies

  • Lectures: Held on Sunday and Tuesday from 10:30 to 12:30 in the Department of Mathematical Sciences, classroom 211.

  • Late Policy: You have a total budget of 15 slack days for the semester, which can be used for any homework (practical or theoretical) without penalty.

    • There is a maximum limit of 5 slack days per assignment. Submissions after 5 days will not be accepted, as solutions may be released.
    • Once your 15-day total budget is exhausted, any further late submission (within the 5-day window) will be penalized 2% of the assignment's grade for every hour of delay.

  • Collaboration & Resources: Collaboration on assignments is permitted. However, your final submission must be written entirely by you. You must clearly cite the names of any collaborators and list all external resources (outside of course materials) that you used.

  • Practical Assignments: Practical (coding) assignments will include an oral defense. You will not receive a grade for the practical portion if you cannot demonstrate sufficient mastery of your code during the defense.

  • Support: You can ask questions on Telegram Group or email the course instructor or head TA for office hours.

Grading

The grading for the Deep Generative Models course is structured as follows:

Assessment Component Points
Homeworks (5 practical & conceptual HWs) 10 + [1 Extra Point]
Midterm 4
Final 6
Total 20 + [1 Extra Point]

Head Assistant

Teaching Assistants