Machine learning for displaced vertex classification

Displaced Vertices are an important signature for new physics searches at the Large Hadron Collider. They point to the existence of long-lived massive particles, which are predicted by many Beyond the Standard Model theories. In this project, we investigate the use of machine learning techniques to classify these displaced vertices in an R-parity violating supersymmetry (SUSY) model, as a simple cut and count can have low signal efficiency. We apply Boosted Decision Trees (BDTs) to secondary vertex information, and Graph Neural Networks (GNN) to fully connected graphs of tracks in the detector. While the BDTs are able to provide an immediate improvement in signal efficiency over the cut and count, the GNNs warrant further investigation.