Machine Learning for Stochastics (SS 2025)

Lecturer: Prof. Dr. Thorsten Schmidt

Lecture: Wed, 12-14 Uhr, HS II, Albertstr. 23b

Tutorial: Monday 12-14, SR 127/128 (Ernst-Zermelo Straße 1)

Language: English

Content

This course introduces modern and highly efficient machine learning tools for solving stochastic problems.

We will cover a range of advanced topics, including neural stochastic differential equations (neural SDEs), which generalize classical SDEs by incorporating neural networks; transformers, not only as powerful models for language but also for time series; and generative models such as GANs for time series generation. Throughout the course, we will emphasize practical applications in Finance and Insurance, including (robust) deep hedging, and the use of reinforcement learning techniques.

Lecture notes / Course materials

An overview of the main topics:

Reinforcement learning and Markov decision processes
Prior fitted networks
Neural and learned SDEs
Generative approaches (e.g., GANs, transformers)
Machine learning for Bayesian inference
Uncertainty quantification
Safe and robust AI methods
Universal approximation theorems
Applications to Finance and Insurance

Exercise sheets

To pass the course (pass/fail examination), you must:

Achieve at least 50% of the maximum possible points from the exercise sheets.
Successfully complete and present a project, where you are expected to implement one of the techniques covered during the course.
Projects may also explore methods not discussed in class, based on literature suggested by the assistant and the lecturer.

You can find the exercise sheets here: Exercise Sheets Repository .

Final exam

The final examination will be an oral exam.

Consultation hours

Lecturer: Available by appointment. Please contact via email.
Assistant: Available anytime by appointment.

For shorter questions, you can also send an email to simone.pavarana@stochastik.uni-freiburg.de