Materials Data Science and Informatics

Content

This module focuses on methods and practical implementations for using data science approaches to extract new information and knowledge from situations of mate- rials scientific relevance. Used methods ranges from dimensionality reduction methods through various deep learning architectures (drop-out, convolutional networks, autoencoder, recurrent and generative adversarial networks) to approaches for high- throughput data analyze (e.g., image data from microscopy) or the use of tailored machine learning methods for predicting materials with new properties. Aspects of practical relevance, e.g. concepts such as batch training, momentum, data augmentation will be discussed and used in hands-on implementations.
The second part of this module is dedicated to the field of materials informatics and introduces concepts relevant for the digitization of materials science as well as for research data management (e.g., the concepts of metadata, ontologies, knowledge- graphs as well as semantic web technologies and the FAIR data principle). This will then be put into the context of accelerated materials design as well as data-driven materials development.

Objective

Students will become familiar with major Deep Learning architectures and under- stand their respective advantage and disadvantages. They are able to analyze a problem of materials scientific relevance and can judge which network architecture and learning approach is the most suitable one. They are able solve problems using open source deep learning libraries and by independently creating their own problem-specific software implementation. Students will be able to judge the quality of their data science results and validate them based on their domain knowledge.
In the second part of this module, students will be exposed to fundamental concepts from Information science as well as from research data management. In particular they will understand the basic concepts of meta data schemas, vocabularies, ontologies and knowledge graphs and learn to apply this to a number of chosen examples. At the end of this module students will be able to understand, analyze and critically judge state-of-the-art topics of data-driven materials science, e.g., from current publications.

basic python programming knowledge; basic knowledge of concepts of statistical learning, e.g., regression, classifications

Charu C. Aggarwal, Neural Networks and Deep Learning: A Textbook 1st ed. 2018 , ISBN 3319944622

Lecture and exercise dates WS23/24

Exercise: Tuesdays 08:30 - 10:00, H08 (C.A.R.L., Claßenstr.11)
Lecture: Tuesdays 10:30 - 12:00, H08 (C.A.R.L., Claßenstr.11)

Exam WS23/24

March 13 2024 14:30 - 16:30, Room 131 main building (Templergraben 55)

Contacts

Administrative Contact: Dr. Katharina Immel
Content: Prof. Dr. Stefan Sandfeld, Dr. Bashir Kazimi

Last Modified: 08.02.2024