40 MHz Scouting with Deep Learning in CMS

This document describes a proof-of-concept study on the possibility of using a neural network to correct the Level-1 (L1) Trigger Muon objects to improve their quality for real-time analysis as part of a 40 MHz scouting system. L1 scouting is the concept of reading out L1 trigger objects at the full bunch crossing rate. This allows for analysis of event types that are too frequent to be part of the nominal L1 menu. The scouting system is independent of the L1 trigger, in that it does not feed back to the trigger decision. A semi-real time analysis using the L1 trigger objects could be possible, or the storage of a tiny event record. The capture of L1 trigger data at 40 MHz was first demonstrated for the Global Muon Trigger (GMT) in 2018, and has been scaled up to include the Layer 2 Calorimeter Trigger for Run 3.