·
Point-Cloud Deep Learning for the Digital Hadronic Calorimeter
Particle identification and energy measurement from the binary hits of a digital hadronic calorimeter, with DeepSets and graph attention networks working directly on the point cloud of hits.
Key result
4×
coarser readout, same performance
With a network on the point cloud of hits, pion energy resolution with 2×2 cm² pads is as good as with 1×1 cm² pads, a quarter of the readout channels. The resolution (48.6%/√E stochastic, 7% constant) is within the target of 55%/√E for future hadron calorimeters.
Resources