Speakers
Description
Efficient identification of heavy flavoured jets is essential for the physics programme of a future muon collider, particularly for Higgs and electroweak measurements where final states involving $b$-quarks dominate. We present the work towards the first implementation of a Graph Neural Network (GNN) based heavy flavour tagger for a Muon Collider detector, built on the SALT framework developed by the ATLAS collaboration. This tagger's neural network architecture treats jets and their constituent tracks as interconnected nodes which suits the dense and topologically complex environment expected at a muon collider, and its geometry-independent design enables porting between detector concepts.
We describe the end-to-end pipeline under development. As part of this pipeline, a format converter tool has been developed to convert MuCol software framework data in SALT-processable input. The simulated events are used to train and evaluate the tagger performance. Simulated $\mu \mu \to b\bar{b} / c\bar{c} /q\bar{q}$ events at $\sqrt{s } =3$ TeV are used for a first-pass training and evaluation of the tagger performance. We discuss the conversion tool developed, the ongoing validation programme, and extension to 10 TeV MUSIC samples and BIB-overlaid conditions.
| What category does your poster fit in? | Software & Simulations |
|---|