Speaker
Description
We introduce a Transformer-based regression model for estimating missing transverse momentum (MET) at hadron colliders. MET represents an imbalance in the vector sum of transverse momenta for all reconstructed particles and serves as a proxy for invisible particles such as neutrinos and dark matter candidates. The deep learning model takes all reconstructed particles as input and directly performs regression on MET components. This model must capture the complex relationship among the many particles to infer the genuine MET originating from invisible particles. This is necessary because pileup particles and non-perfect detector responses can contaminate the momentum imbalance in a reconstructed event. Thus, the model deploys the Transformer's attention mechanism, enabling the extraction of long-range dependencies among particles. In this presentation, we will showcase the results of applying the Transformer-based model to the Monte Carlo simulation of top quark pair production in the dilepton channel in pp collisions.
