Advancements in flavour tagging and tracking using ML at ATLAS

Flavour tagging is a crucial component of the ATLAS experiment’s physics programme. Recent advancements harness Graph Neural Networks and Transformers to process low-level detector inputs directly, enabling a streamlined, end-to-end approach. This paradigm shift not only boosts tagging performance—particularly in challenging regimes such as high transverse momentum and c-jet identification—but also enhances interpretability through auxiliary tasks related to vertexing and track classification. By reducing reliance on hand-crafted inputs, these methods mark a significant step forward in jet identification, with applications extending beyond conventional b- and c-tagging to boosted Higgs tagging, searches for long-lived particles, and jet energy calibration. The seminar will also explore advanced object detection techniques, which are paving the way for more powerful track and vertex reconstruction in high-energy physics experiments.

Event resources