Graph Neural Networks for Particle Momentum Estimation in the CMS Trigger System

Description

CMS experiment currently uses machine learning algorithms at the Level-1 (hardware) trigger to estimate the momentum of traversing particles such as Muons. The first algorithm implemented in the trigger system was a discretized boosted decision tree. Currently, CMS is studying the use of deep learning algorithms at the trigger level that requires microsecond level latency and therefore requires highly optimized inference.

This project will focus on implementation and benchmarking of deep learning algorithms for the trigger inference task.

Duration

Total project length: 175/350 hours.

Task ideas

Development and Benchmarking of graph networks (GNN) for momentum regression in the trigger system

Expected results

Benchmarks of deep network model inference for muon momentum assignment for prompt and displaced particles

Requirements

Python, C++, and some previous experience in Machine Learning.

Mentors

Suzanne Rozenzweig (University of Florida)
Efe Yigibasi (University of Florida)
Darin Acosta (University of Florida)
Sergei Gleyzer (University of Alabama)
Ali Hariri (American University Beirut)

Please DO NOT contact mentors directly by email. Questions should instead be directed to ml4-sci@cern.ch which is forwarded to mentors. To submit your proposal, CV, and test task solutions, please use this Google form.