57. Heidelberger Physik Graduiertentage

From basic classification task, to generative simulation and advanced agentic workflows, machine learning methods have revolutionized many aspects of research work. 

This lecture series offers a practical tour of the methods that matter most for working physicists, what these methods do, when to reach for them, and the tricks that make them actually work.

 

Lecture will cover core tools, such as supervised learning, regression and classification, and the surprising range of physics problems they solve. From there we introduce the transformer architectures that now form the backbone of modern ML and AI, before turning to generative models that can simulate data and stand in for expensive experiments or computations. Next, we will cover a core concern for any ML use in physics: uncertainty quantification, and finally we introduce LLMs and agentic systems and demonstrate how to use them in scientific applications. 

 

This lecture series is accessible without prior knowledge of ML, but aims to still provide new insights to students with some prior ML experience.