Long Short-Term Memory (LSTM) networks with jet constituents for boosted top tagging at the LHC

• DOI: 10.48550/arxiv.1711.09059
• Publication Date: 2017-01-01
• Authors: Shannon Egan; Wojciech Fedorko; A. Lister; J. Pearkes; C. Gay
• Keywords: Calorimeter (particle physics); Identification; SIGNAL (programming language)

Multivariate techniques based on engineered features have found wide adoption in the identification of jets resulting from hadronic top decays at Large Hadron Collider (LHC). Recent Deep Learning developments this area include treatment calorimeter activation as an image or supplying a list jet constituent momenta to fully connected network. This latter approach lends itself well use Recurrent Neural Networks. In work applicability architectures incorporating Long Short-Term Memory (LSTM) networks is explored. Several network architectures, methods ordering constituents, and input pre-processing are studied. The best performing LSTM achieves background rejection 100 for 50% signal efficiency. represents more than factor two improvement over Network (DNN) trained similar types inputs.