Seminar: Generative AI with Diffusion Techniques

Administrative information

Seminar course for Bachelor students (IN0014) and Master students (IN2107).

  • Organizer: Tobias Lasser
  • Sessions: the course takes place online.
    • Block 1+2: basics and diffusion techniques
      • Wednesday, October 8, 2025, 13:30 - 17:30
      • Thursday, October 9, 2025, 13:30 - 17:30
    • Block 3+4: optimization techniques and applications
      • Wednesday, December 10, 2025, 13:30 - 17:30
      • Thursday, December 11, 2025, 13:30 - 17:30
  • Course language: English

Pre-course meeting

A pre-course meeting will take place on Wednesday, July 16, 2025, at 11:00 online (BBB room).

Registration

  • Registration takes place via the IN.TUM Matching System from July 18 to 22, 2025.
  • Additionally, please fill out the registration form until July 22, 2025.
  • If you do not fill out the registration form, you will not be matched to the course!

The seminar is open to all students of Informatics, Mathematics, Physics and related fields.
The number of students is limited to 8.

Course overview

Generative AI based on diffusion models is one of the most dynamic and rapidly evolving areas in both academic research and real-world applications. Today's most exciting use cases include generating images (e.g., text-to-image with tools like Stable Diffusion or DALL·E) and videos (e.g., text-to-video with models like Veo 3 or Cosmos). In this seminar, you'll explore the foundations of diffusion-based generative techniques, focusing on how they power modern image and video generation. We will also dive into optimization strategies that reduce inference costs and explore how these techniques extend beyond media to impactful domains such as medical imaging and genomics.

As part of this course, you will actively contribute to the learning experience by giving two in-depth presentations (30 minutes + 15 minutes discussion) based on cutting-edge scientific literature. The first presentation (Blocks 1 & 2 in October) will cover foundational concepts. The second presentation (Blocks 3 & 4 in December) will build on this base, diving deeper into specialized or applied topics. Interactive sessions and audience engagement are highly encouraged to foster a collaborative learning environment. To wrap up the seminar, each student will submit a concise 3-page written summary covering the key topics and discussions, creating a lasting reference for all participants. Whether you're interested in the theory behind diffusion models, their practical implementation, or their broader impact across disciplines—this seminar offers a hands-on, research-driven experience at the forefront of AI.

Aims of the course

This seminar aims to equip students with a solid understanding of diffusion-based generative AI techniques, both from a theoretical and practical perspective. Through in-depth exploration of current research, students will gain insight into how these models are used to generate images, videos, and domain-specific data. The course fosters critical thinking, presentation skills, and collaborative discussion, while encouraging students to engage with cutting-edge developments in AI and explore innovative applications across disciplines.

Prerequisites

Participants should have a basic understanding of deep learning concepts. A strong interest in generative AI and enthusiasm for the interactive, presentation-based seminar format are highly recommended.