Deep Neural Network MET Reconstruction for the CMS Level-1 Trigger

In anticipation of the high-luminosity phase of the Large Hadron Collider (LHC), the Compact Muon Solenoid (CMS) collaboration presents advancements in event selection at the Level-1 (L1) Trigger. The projected ten-fold increase in luminosity introduces challenges such as higher data volumes and increased pile-up complexity. Missing transverse energy (MET), a key observable at CMS, is crucial for both event reconstruction and as a potential signature of exotic particles. L1DeepMET is an adaptation of the DeepMET neural network, previously used in offline MET reconstruction, optimized for integration into the L1 Trigger. We present the performance of L1DeepMET after compression and quantization, ensuring it meets the stringent sub-microsecond latency requirements of the L1 Trigger. The model features two dense layers and processes up to 100 ParticleFlow candidates per event. Our results demonstrate that L1DeepMET achieves performance comparable to the original DeepMET model, while being fully compatible with the real-time constraints of the L1 trigger.