Graph Neural Network for Real-Time Top Quark Tagging

We present a Graph Neural Network-based approach to the problem of tagging physics events as originating from either a hadronically-decaying top quark or from QCD processes. We use Monte Carlo simulated samples at a center-of-mass energy of 14 TeV. These are converted to a graph representation in which the reconstructed quantities from the event are used as node features. We train a GraphSAGE-based Graph Neural Network to classify these graphs using MATLAB, and compare its performance with other algorithms. Finally, we discuss the potential of the MATLAB implementation to be applied directly to hardware for real-time trigger purposes.