Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Advancing the CMS Level-1 Trigger: Jet Tagging with DeepSets at the HL-LHC

Stella Schaefer, Christopher Brown, Duc Hoang, Sioni Summers, Sebastian Wuchterl

TL;DR

The paper addresses the challenge of selecting rare processes at the HL-LHC with the CMS Phase-2 Level-1 Trigger by introducing a small, quantized Deep Sets-based multiclass jet tagger operating on Seeded Cone jets to classify eight jet categories, and it demonstrates its performance and FPGA feasibility. The approach includes a 16-constituent input, a $p_T$ regression branch, and an FPGA-friendly architecture designed for strict L1T latency and resource limits. Results show robust discrimination across jet flavors, significant seed-efficiency gains at low $H_T$, and a practical hardware implementation: latency ~1 μs, 360 MHz throughput, and modest FPGA resource usage, with Seed studies indicating tangible trigger improvements for di-Higgs channels. Overall, the work provides a viable path to richer jet flavor tagging at L1T, enabling improved trigger selections under extreme pileup.

Abstract

At the High Luminosity LHC, selecting important physics processes such as (di-) Higgs production will be a high priority. The Phase-2 Upgrade of the CMS Level-1 Trigger will reconstruct particle candidates and use pileup mitigation for the 200 simultaneous proton-proton interactions. A fast cone algorithm will reconstruct jets from these particles, providing access to jet constituents for the first time. We introduce a new multi-class jet tagger with a small, quantized DeepSets neural network. The tagger, trained on a mix of simulated CMS events, predicts various hadronic and leptonic classes. We present the tagger, its performance, and its improvements for triggering on (di-) Higgs events.

Advancing the CMS Level-1 Trigger: Jet Tagging with DeepSets at the HL-LHC

TL;DR

The paper addresses the challenge of selecting rare processes at the HL-LHC with the CMS Phase-2 Level-1 Trigger by introducing a small, quantized Deep Sets-based multiclass jet tagger operating on Seeded Cone jets to classify eight jet categories, and it demonstrates its performance and FPGA feasibility. The approach includes a 16-constituent input, a regression branch, and an FPGA-friendly architecture designed for strict L1T latency and resource limits. Results show robust discrimination across jet flavors, significant seed-efficiency gains at low , and a practical hardware implementation: latency ~1 μs, 360 MHz throughput, and modest FPGA resource usage, with Seed studies indicating tangible trigger improvements for di-Higgs channels. Overall, the work provides a viable path to richer jet flavor tagging at L1T, enabling improved trigger selections under extreme pileup.

Abstract

At the High Luminosity LHC, selecting important physics processes such as (di-) Higgs production will be a high priority. The Phase-2 Upgrade of the CMS Level-1 Trigger will reconstruct particle candidates and use pileup mitigation for the 200 simultaneous proton-proton interactions. A fast cone algorithm will reconstruct jets from these particles, providing access to jet constituents for the first time. We introduce a new multi-class jet tagger with a small, quantized DeepSets neural network. The tagger, trained on a mix of simulated CMS events, predicts various hadronic and leptonic classes. We present the tagger, its performance, and its improvements for triggering on (di-) Higgs events.

Paper Structure

This paper contains 7 sections, 3 figures.

Table of Contents

  1. Introduction
  2. Tagging strategy
  3. Network architecture
  4. Jet tagging performance
  5. Seed studies
  6. FPGA implementation
  7. Conclusion

Figures (3)

  • Figure 1: Deep Sets architecture for jet tagging and $p_T$ regression. All 16 constituents are processed individually before aggregating them through a global average and splitting into a $p_T$ regression branch and a jet ID branch dps.
  • Figure 2: One-vs-one ROC curves and AUC scores evaluated on a $t\Bar{t}$ physics sample dps.
  • Figure 3: Comparison of a purely kinematic $H_T$ seed and tagging + $H_T$ as well as a comparison to previous b-jet tagging developments for the HH $\rightarrow$ 4b process. The new seed shows significant efficiency gain at low $H_T$dps.