Machine learning (ML) is having a profound impact on physics research. ML is now ubiquitous, and is enabling previously-intractable computational problems to be framed in new ways and solved to the benefit of scientific research. While ML methods have been used in high energy physics (HEP) for many years, rapid progress over the last decade including the development of deep learning and continuously evolving computing architectures, has led to a revolution in this area.
Geometric Deep Learning (GDL) describes a class of ML algorithms that are capable of learning from a range of geometric data types including graphs, point clouds, manifolds, and sets. In particular, these algorithms have demonstrated a capacity to learn information about inherent geometric structures and symmetries. Graph Neural Networks (GNNs) are a powerful class of GDL methods for modeling spatial dependencies via message passing over graphs. They are increasingly being applied to overcome key computational challenges in particle physics and are guiding new approaches to data-driven discovery in exciting ways.
In this talk, I will discuss a few key applications of GNNs to particle physics with an emphasis on the challenge of track reconstruction during the High-Luminosity Large Hadron Collider (HL-LHC) era. The HL-LHC will provide an order of magnitude increase in integrated luminosity as compared with current running and enhance the discovery reach for new phenomena. The associated increased pile-up foreseen during the HL-LHC necessitates major upgrades to the ATLAS detector and trigger. GNNs are well-suited for track reconstruction tasks by learning on an expressive structured graph representation of hit data in tracking systems and provide an opportunity for considerable speedup over CPU-based execution using accelerator hardware such as FPGAs.
