Deep-Neural-Network-based b-Tagging as Basis for Improvements in Top Analyses

Analyses involving top quarks are characterised by the presence of several b-jets in the final state. An efficient discrimination between b- and non-b-jets is crucial in order to select the signal processes and reject the physics background. Developments in the field of machine-learning in the recent years allow to design more sophisticated b-tagging algorithms. Those improvements are hugely beneficial for top analyses. One of those b-tagging algorithms is the deep-learning based high-level tagger (DL1) in ATLAS. Studies applying these algorithms to the recently introduced particle-flow jets are presented, leading to a large improvement in the b-jet identification.