Seminars

Deep learning has been successfully applied in a wide range of use cases and specifically in applications involving visual and textual data. Modern machine translation systems and search engines, for example, are using language models trained on large text corpora. Increasingly, deep learning is also applied to problems arising in the sciences and engineering. For instance, deep learning for graphs is used to learn simulators from data. The figure on the right shows a particle simulation obtained from a graph neural network trained on simulation traces from a numerical solver. The advantage of using deep learning in this context is its ability to integrate the simulation into a larger neural network with a corresponding loss function and train the resulting model end-to-end.

GS SimTech-Members, Participants of the seminar are expected to have completed the course “Machine Learning", "Reinforcement Learning", or a related course.

Depending on the number of participants, the exam will be in written or oral form.

Students who do not take the exam but need a proof of attendance (e.g. PhD students, FÜSQ participants) have to present their results for at least one minor and one major exercise in the Inverted Classroom meetings.

How do I get to the heart of my topic and how do I communicate it to the public? This workshop series offers a DeepDive into the most important techniques of science communication: we look for and find the topic in our subject, we find best sentences, write a pitch, practice repartee for interview situations and learn how a gripping speech is structured. In the end, everyone will have something in hand: a self-written press release, a speech concept, a poster presentation or an artistic format. And we will have a lot of fun.

Depending on the number of participants, the exam will be in written or oral form.

Students who do not take the exam but need a proof of attendance (e.g. PhD students, FÜSQ participants) have to present their results for at least one minor and one major exercise in the Inverted Classroom meetings.

This novel course addresses master and PhD students with basic knowledge and experience in continuum mechanics based mathematical modeling and numerical simulations of natural or technical processes. Its goal is to provide or refresh or widen the necessary background information from mathematics, both concerning model formulation and validity and discretization and solution methods. Due to this wide span, the lecture can only touch selected topics, also relying on the active reading activity of the participants. In particular, specific interests and questions of the participants can be addressed in the discussion hours. Examples of questions which may come up and be discussed are

• Is my model thermodynamically consistent ?
• What means mathematically rigorous or formal, what are the consequences ?
• What am I allowed to do with which functions ?
• What is a good discretization method ?
• Is my discretization locally mass conservative, if not, what to do?
• Etc.