Course Information
Description
This seminar is designed for master students of Mathematics of Machine Learning and Data Science, Scientific Computing and Mathematics. We consider Machine Learning from a mathematical point of view and discuss its relation to new as well as well established mathematical concepts. This semester the focus topic is Generative Models and Measure Transport. Together we will explore various aspects of five selected papers, e.g. content, authors, references, related research, etc.
All five papers address the same problem: samples from an unknown distribution are given, and new ones are to be generated. They also solve it in the same way, by pushing a simple reference measure onto the data distribution. What separates the five papers is the notion of closeness they commit to.
The seminar style will be similar to what is described in: https://colinraffel.com/blog/role-playing-seminar.html We have selected five roles. For more details on the roles we refer to the preparatory meeting. Each participant has been assigned a number and fills every role exactly once over the five sessions. As there are seven participants, each session has two roles filled by a team of two, with the remaining three taken individually.
The assignment by paper and role:
The same assignment from each participant’s perspective:
Papers
In every meeting we will study together one particular paper on generative models and measure transport.
- Kingma, Welling: Auto-encoding variational Bayes
- Goodfellow, Pouget-Abadie, Mirza, Xu, Warde-Farley, Ozair, Courville, Bengio: Generative adversarial nets
- Arjovsky, Chintala, Bottou: Wasserstein generative adversarial networks
- Ho, Jain, Abbeel: Denoising diffusion probabilistic models
- Song, Sohl-Dickstein, Kingma, Kumar, Ermon, Poole: Score-based generative modeling through stochastic differential equations
Exam / Presentations
For a successful completion of the seminar it is mandatory to attend all sessions. In the beginning of every meeting each participant gives a five minute report from the perspective of their role, followed by a joint scientific discussion of the paper. The final grade will be a weighted sum of the grades of your presentations and of your participation in the subsequent discussions.
Prerequisites
No prior knowledge of generative models is required. Familiarity with measure theoretic probability is helpful.
Dates and Timeline
- Organizational meeting:
- October 13th, 2026, SR 10 of INF 205.
- Participation in this meeting is essential to attending the seminar.
- Roles will be assigned at this meeting.
Meetings will be held on the following Tuesdays from 16-18h at SR 10 of INF 205:
- 27.10.2026
- 10.11.2026
- 24.11.2026
- 08.12.2026
- 12.01.2027
(Pre-)registration
Registration is handled via MaMpf . The seminar is limited to 10 participants; this semester seven are registered.