Modulcode: | infSemAMLMe-01a |
Englische Bezeichnung: | Master Seminar - Advances in ML Methods in Materials Science and Engineering |
Modulverantwortliche(r): | Dr. Roland Aydin |
Turnus: | unregelmäßig (SS22) |
Präsenzzeiten: | 2S |
ECTS: | 5 |
Workload: | 30 Std. Mitarbeit im Seminar, 120 Std. Selbststudium |
Dauer: | ein Semester |
Modulkategorien: | MSc-Inf-Sem (MSc Inf (21)) Sem (MSc Inf (15)) |
Lehrsprache: | Englisch |
Voraussetzungen: |
In the evolving field of Machine Learning, in particular for the subdomain of artificial neural networks, many advances have been proposed in recent years. In this seminar, students choose from a pool of publications with each describing a new state-of-the-art AI method relevant for materials science and engineering. The participants then get to know the chosen method in-depth and present their new-found knowledge to the other seminar participants in a format similar to a scientific conference. They also provide a critical review both of the method they themselves presented, as well as in shorter form of the other methods presented by their fellow students. A second part of the seminar then includes a hands-on implementation of a proof-of-concept version of the new method in Python.
Students attain both theoretical and practical competency in a number of recent Machine Learning advances. They also learn how to scientifically present, evaluate, and critique both their chosen as well as other current AI methods.
The seminar aims to convey a broad sense of recent developments in methods based on artificial neural networks in general, as well as a specific and thorough theoretical and practical understanding of one new approach (chosen by the student from a pool of such methods) in particular. The set of subjects is selected as being of particular importance to the fields of Material Science and Engineering, but with potentially broader applicability to other Machine Learning subfields. Examples include automatic topology optimization, multi-fidelity simulations for generating training data, feature selection in a highly multidimensional parameter space, and similar. The practical component of then implementing a simplified version of the method will be tailored to the individual level of Python expertise of each student, while retaining the salient features demonstrating the method's usability and thus the student's understanding.
Prior knowledge in machine learning (basics) and artificial neural networks. Prior basic knowledge in a ML library in Python.
The grade is composed of the following aspects:
The students will be given specific literature to work on in the seminar (changes from term to term).