Prof. Gregor Kasieczka

T a l k : 8. May 2025

Towards AI-based Foundation Models for Physics Research

Machine learning algorithms are gradually replacing traditional rule-based systems in various parts of modern physical research. However, these models are generally tailored to perform a specific task on a well-defined dataset. The concept of foundation models, particularly those based on transformer architectures, offers an alternative approach.  These models are designed with enough complexity to be trained on diverse tasks and datasets, allowing them to be efficiently applied or fine-tuned for new problems.
In this seminar we will first consider the current state of proto foundation models for data-analysis applications in particle physics. We then explore the requirements needed to go beyond even such models towards a potential multi-domain physics AI system.