| Kommentar |
Data-driven techniques have become more and more important in the field of materials. The newly emerging field has been termed materials informatics. This course will introduce the materials informatics to students in physics. Interest in these topics has also grown significantly on the industry side in recent years, as showcased by major publications of important AI-focused companies.
This field of materials informatics is closely linked to computational physics and materials science. Data analysis and machine learning techniques are used, but these are specifically tailored to materials. In addition, understanding material data and its origin from both (atomistic) simulations and experiments plays a significant role.
At the end of the course, all students should be able to work on materials informatics topics in practice. The basis for the lectures and exercises will be the programming language Python. They will rely on frequently used open-source codes in the field (https://pymatgen.org/, https://wiki.fysik.dtu.dk/ase/, https://matminer.readthedocs.io/en/latest/, https://scikit-learn.org/stable/, https://pytorch.org/ ). The programming exercises in (object-oriented) Python are also expected to expand the students' programming skills significantly.
The following topics will be covered:
- Object-oriented programming and data science with Python (including usage of Pandas), Introduction to git
- Data sources and access to material data (e.g., https://next-gen.materialsproject.org/ or https://nomad-lab.eu/nomad-lab/)
- Automation of data generation (e.g., using density functional theory or machine-learned interatomic potentials)
- Typical descriptors for materials (representation of the composition of crystalline or amorphous solids or the structure of crystalline solids)
- General principles of machine learning
- Classification and regression
- Supervised and unsupervised learning
- Clustering
- Kernel methods
- Neural networks (different architectures)
- current examples from materials informatics (e.g., https://doi.org/10.1038/s41586-023-06735-9, https://arxiv.org/pdf/2312.03687.pdf)
Examination type:
Homework project plus presentation of the project. |
| Literatur |
As the field of materials informatics is very new, no dedicated textbook for materials informatics exists. Further general resources on data science, machine learning and electronic structure theory will be provided during the course. In addition, detailed material will be provided as a part of the lecture.
Inspiration for the course has been based on the following lectures, as provided on github:
https://github.com/sp8rks/MaterialsInformatics/
https://github.com/enze-chen/mi-book-2021 |
