Probabilistic Diffusion: Theory and Applications

Summer Term 2023

Seminar: Probabilistic Diffusion: Theory and Applications

Karl Schrader, Kristina Schaefer, Prof. Joachim Weickert

Summer Term 2023

Seminar (2 h)

Cats. (Authors: Midjourney)


February 20, 2022:
The website is online.

Important DatesDescriptionRegistrationRequirementsIntroductory MeetingOverview of TopicsSupplementary Material

Introductory meeting (mandatory):
The introductory meeting will take place in building E1.7, room 4.10. on Thursday, April 20, 2023, 14:15. In this meeting, we will assign the topics to the participants.
Attendance is mandatory for all participants. Do not forget to register first (see below).

Regular meetings online during the summer term 2023:
Thursday, 14:15 - 16:00 in building E1.7, room 4.10.

Write-up submission deadline:

Contents: Probabilistic diffusion models are a popular class of neural networks, most famous for their ability to generate realistic images from text descriptions. They are a point of fascination not only for reasearchers, but have become a society-wide phenomenon. This seminar will explore the principles behind probabilistic diffusion models and will span the whole range from stochastic equations, over Markov processes, to beautiful images. Through our list of topics, participants will learn about the mathematical foundations of those models, and which insights are needed to turn those theoretical results into practice. We will explore the differences between popular models, and how their training can be improved. Furthermore, we will lightly touch upon adjacent areas like probablistic diffusion for audio generation, or ideas to protect original art styles from being mimicked by AI.

Prerequisites: The seminar is for advanced bachelor or master students in Visual Computing, Mathematics, or Computer Science. Basic mathematical knowledge (e.g. Mathematik für Informatiker I-III) and some knowledge in image processing and computer vision is required.

Language: All papers are written in English, and English is the language of presentation.

Mode: The seminar will be conducted in person. We will provide a Team on the MS Teams platform for communication with your advisor and submission of the write-up.

You must register for this course via the Seminar System

from March 13, 2022
until April 12, 2022

Regular attendance: You must attend all seminar meetings. If you are sick, please send a medical certificate via mail to Karl Schrader.

Talk: Talk duration is 30 min, plus 15 min for discussion. Please do not deviate from this time schedule. More details and a best practice guide will be made available after the introductory meeting.

Write-up: The write-up has to be handed in three weeks after the lecture period ends. The deadline is 13.08.2023. The write-up should summarise your talk and has to consist of 5 pages per speaker. Please adhere to the guidelines for write-ups posted in our Teams group. Mail your write-up in pdf format directly to your advisor.

Plagiarism: Adhere to the standards of scientific referencing and avoid plagiarism: Quotations and copied material (such as images) must be clearly marked as such, and a bibliography is required. Otherwise the seminar counts as failed. See the write-up guidelines for a detailed explanation on how to cite correctly.

Mandatory consultation: Talk preparation has to be presented to your seminar supervisor no later than one week before the talk is given. It is your responsibility to approach us timely and make your appointment for a video call or a personal meeting.

No-shows: No-shows are unfair to your fellow students: Some talks are based on previous talks, and your seminar place might have prevented the participation of another student. Thus, in case you do not appear to your scheduled talk (except for reasons beyond your control), we reserve the right to exclude you from future seminars of our group.

Participation in discussions: The discussions after the presentations are a vital part of this seminar. This means that the audience (i.e. all paricipants) poses questions and tries to find positive and negative aspects of the proposed idea.

Being on time: All participants have to be in the seminar meeting on time. Participants that turn out to be regularly late must expect a negative influence on their grade.

The slides of the introductory meeting will be uploaded in Teams. They contain important information for preparing a good talk.

As the research around probablistic diffusion models is ongoing and progressing rapidly, we might still make minor changes to the list of papers.

No.   Date   Topic
1 11.5. N. Carlini, J. Hayes, M. Nasr, M. Jagielski et al.:
Extracting Training Data from Diffusion Models
2 11.5. S. Shan, J.Cryan, E. Wenger, H. Zheng et al.:
GLAZE: Protecting Artists from Style Mimicry by Text-to-Image Models
3 1.6. J. Sohl-Dickstein, E. A. Weiss, N. Maheswaranathan, and S. Ganguli:
Deep Unsupervised Learning using Nonequilibrium Thermodynamics
4 1.6. Y. Song and S. Ermon:
Generative Modeling by Estimating Gradients of the Data Distribution
5 15.6. I Kobyzev, S. J. D. Prince, and M. A. Brubaker:
Normalizing Flows: An Introduction and Review of Current Methods
6 15.6. J. Ho, A. Jain, and P. Abbeel:
Denoising Diffusion Probabilistic Models
7 22.6. Y. Song, C. Durkan, I. Murray, and S. Ermon:
Maximum Likelihood Training of Score-Based Diffusion Models
8 22.6. D. P. Kingma, T. Salimans, B. Poole, and J. Ho:
Variational Diffusion Models
9 6.7. J. Ho and T. Salimans:
Classifier-Free Diffusion Guidance
10 6.7. A. Radford, J. W. Kim, C. Hallacy, A. Ramesh et al.:
Learning Transferable Visual Models From Natural Language Supervision
11 13.7. A. Nichol, P. Dhariwal, A. Ramesh, P. Shyam et al.:
GLIDE: Towards Photorealistic Image Generation and Editing with Text-Guided Diffusion Models
12 13.7. R. Rombach, A. Blattmann, D. Lorenz, P. Esser et al.:
High-Resolution Image Synthesis with Latent Diffusion Models
13 20.7. T. Salimans and J. Ho:
Progressive Distillation for Fast Sampling of Diffusion Models
14 20.7. Z. Kong, W. Ping, J. Huang, K. Zhao et al.:
DiffWave: A Versatile Diffusion Model for Audio Synthesis

Description Resource
A very readable intro to diffusion models. S. Karagiannakos and N. Adaloglou:
How diffusion models work: the math from scratch
A step-by-step code explanation of paper 6. N. Rogge and K. Rasul:
The Annotated Diffusion Model

MIA Group
The author is not
responsible for
the content of
external pages.

Imprint - Data protection