Machine learning for track reconstruction at the LHC

Abstract

The planned upgrade of the LHC to its High-Luminosity counterpart (HL-LHC) circa 2027 will bring about a drastic increase in instantaneous luminosity, pileup, and trigger rates. Currently, most LHC experiments use Kalman filter based track reconstruction algorithms which exhibit outstanding physics performance but scale poorly with the amount of data produced per bunch crossing. Therefore, the high energy physics community is currently performing intensive R&D to commission new or improved algorithms for this crucial data reconstruction task. This article presents many approaches such as running existing Kalman filter algorithms on accelerated hardware and overhauling the current approaches with machine learning techniques. A new algorithm testbed for research in track reconstruction, ACTS, is also discussed.