The flavour tagging, i.e. the identification of jets originating from heavy flavour quarks, is an essential task for the Standard Model and Beyond the Standard Model research at colliders. Machine Learning-based algorithms have been playing a key role since long time in this task. Graph Neural Networks (GNNs) are a type of machine learning tool where input datasets are represented and processed as graphs. In the context of flavour tagging, GNNs can be particularly useful as they can represent and exploit the internal structure of jets for the identification of the original parton by utilizing the tracks associated with the jets.
In this article, we present AUTOGRAPH (Automatic Unified Training and Optimization for Graph Recognition and Analysis with Pipeline Handling), a fully automated and totally customizable pipeline based on GNNs dedicated to flavour tagging.