Introduction into various approaches and methods for using Machine Learning in conjunction with physical systems. \n
Topics include:
- Data pre- and postprocessing, classification versus regression
- Decision-trees and random forests
- Convolutional Neural Networks (CNNs)
- Feature selection
- Neural architecture search (NAS) and hyperparameter tuning
- Constitutive artificial neural networks (CANNs)
- Synthetic data
- Multimodal and ensemble learning
- Optimal experimental design (active learning)
- Large Language Models
- Process-structure-properties machine learning pipelines
All these methods are useful in non-physical domains as well, the focus of the lecture and exercise will be their usability for physical systems.
The associated exercise sessions (on the same day) will make use of various Python-libraries such as Sklearn and Pytorch, usually using Jupyter notebooks. Knowledge from the exercises will be relevant for the lecture and vice versa.
No prior knowledge in machine learning or Python programming is strictly required, although it would be beneficial.
Performance accreditation:
m1807-2022 - Machine Learning for Physical Systems<ul><li>p1888-2023 - Machine Learning for Physical Systems: Klausur schriftlich</li></ul>
ECTS credit points:
3
Stud.IP informationen about this course:
Home institute: Machine Learning in Virtual Materials Design (M-EXK5)
Registered participants in Stud.IP: 86
Postings: 3
Documents: 15