TU-01: Basic Introduction to Machine Learning in a Physical Context

Real-world modelling is a fundamental concern in physics. The use of self-learning methods, machine learning (ML), is the method of choice when dealing with large data sets, inverse problems, complex feature spaces and more. However, models are abstractions of the real world and thus implicitly subject to uncertainty. The applicability of ML in physics requires a basic understanding of the limitations of the applied ML-model, i.e. its trustworthiness and more precisely its dependability. This tutorial aims to provide an introduction to ML, assuming no prior knowledge beyond basic scientific understanding. An overview of ML tasks, models and data types relevant to applications in magnetism is given. A hands-on part will introduce the basic concept of a ML-based data evaluation approach using open-source software and data sets. The audience is asked to bring their laptops and follow along. The last part of the tutorial will cover the topic of ML dependability. The basic elements of dependability assessment of ML models will be introduced with a special focus on robustness.