The effect of Neural Networks on hadron calorimetry with Cherenkov fiber calorimeters

Improving the reconstruction of hadronic shower with similar precision as electromagnetic particles in high-energy physics calorimeters has been a longstanding quest. The dual-readout method (DREAM) measuring the scintillation and Cherenkov light simultaneously has shown the significant improvement of the energy reconstruction for hadronic shower over traditional simple signal sum method. We simulated a finely-segmented Cherenkov fiber calorimeter using GEANT4 and used neural network to reconstruct the energy. We compare its performance to the dual-readout method and discuss how the neural networks (graph neural network) improves the energy reconstruction of hadronic shower.