Regression Deep Neural Networks for top-quark-pair resonance searches in the dilepton channel

Several Beyond Standard Model (BSM) theories predict the existence of new massive particles decaying to pairs of top quarks, $t\bar{t}$. In this concept work, the key observable for such resonance searches, the top-pair system invariant mass, $m_{t\bar{t}}$, is reconstructed by training a deep neural network on a sample of simulated $t\bar{t}$ events. A regression task is then performed on both $t\bar{t}$ and $Z^\prime$ signal events, using $m_{t\bar{t}}$ as output parameter. The comparison between this machine learning approach and more traditional system reconstruction techniques highlights a tangible improvement in the ability to correctly reconstruct and resolve a TeV-scale $t\bar{t}$ resonance peak.