MLPF: efficient machine-learned particle-flow reconstruction using graph neural networks

Joosep Pata,Javier Duarte,Maurizio Pierini,Jean-Roch Vlimant,Maria Spiropulu

doi:10.1140/epjc/s10052-021-09158-w

Abstract

In general-purpose particle detectors, the particle-flow algorithm may be used to reconstruct a comprehensive particle-level view of the event by combining information from the calorimeters and the trackers, significantly improving the detector resolution for jets and the missing transverse momentum. In view of the planned high-luminosity upgrade of the CERN Large Hadron Collider (LHC), it is necessary to revisit existing reconstruction algorithms and ensure that both the physics and computational performance are sufficient in an environment with many simultaneous proton–proton interactions (pileup). Machine learning may offer a prospect for computationally efficient event reconstruction that is well-suited to heterogeneous computing platforms, while significantly improving the reconstruction quality over rule-based algorithms for granular detectors. We introduce MLPF, a novel, end-to-end trainable, machine-learned particle-flow algorithm based on parallelizable, computationally efficient, and scalable graph neural network optimized using a multi-task objective on simulated events. We report the physics and computational performance of the MLPF algorithm on a Monte Carlo dataset of top quark–antiquark pairs produced in proton–proton collisions in conditions similar to those expected for the high-luminosity LHC. The MLPF algorithm improves the physics response with respect to a rule-based benchmark algorithm and demonstrates computationally scalable particle-flow reconstruction in a high-pileup environment.

Highlights

Reconstruction algorithms at general-purpose high-energy particle detectors aim to provide a holistic, well-calibrated physics interpretation of the collision event
We report the physics and computational performance of the machine-learned particle-flow (MLPF) algorithm on a Monte Carlo dataset of top quark–antiquark pairs produced in proton–proton collisions in conditions similar to those expected for the highluminosity Large Hadron Collider (LHC)
We find that the model generalizes well to the QCD sample that was not used in the training, demonstrating that the MLPF-based reconstruction is transferable across different physics samples

Summary

Introduction

Reconstruction algorithms at general-purpose high-energy particle detectors aim to provide a holistic, well-calibrated physics interpretation of the collision event. [40], important physics effects such as pair production, various ML models including GNNs and computer-vision Brehmsstrahlung, nuclear interactions, electromagnetic showmodels have been studied for reconstructing neutral hadrons ering or a detailed detector simulation, it allows the study of from multi-layered granular calorimeter images and track- overall per-particle reconstruction properties for charged and ing information. The inputs to PF are charged particle tracks and calorimestruction approaches by extending the ML-based PF algo- ter clusters We use these high-level detector inputs (elerithm to reconstruct particle candidates in events with a large ments), rather than low-level tracker hits or unclustered number of simultaneous pileup (PU) collisions. In future iterations of MLPF, it may be beneficial to represent input elements of different types with separate data matrices to improve the computational efficiency of the model Precomputing additional features such as track trajectory intersection points with the calorimeters may further improve the performance of PF reconstruction based on machine learning. We approximate the reconstruction as set-to-set translation and implement a baseline MLPF reconstruction using GNNs

ML-based PF reconstruction

Results

Discussion and outlook