Speaker
Description
The LHCf experiment measures forward neutral particle production at the LHC, providing key inputs for the tuning of hadronic interaction models used in ultra-high-energy cosmic ray physics. The reconstruction of multi-photon final states in forward experimets represents a challenging offline computing problem, due to overlapping showers, non-uniform detector response, and strong correlations between reconstructed objects. In the LHCf experiment, this issue is particularly relevant for events associated with neutral kaon ($K^0_s$) production, which typically result in four-photon signatures in the Arm2 detector.
In this contribution, we present a machine learning-based offline reconstruction pipeline developed for $K^0_s$ candidate events in LHCf. The pipeline employs a neural network classifier to identify events containing four photons, achieving an efficiency of approximately 75-80$\%$ on simulated data. This selection is used as input to a dedicated regression network that performs simultaneous energy $E$ and position ($x,y$) reconstruction for each photon in the event.
The model processes heterogeneous detector inputs, including energy deposits from the two Arm2 calorimetric towers and signals from the silicon strip detectors oriented along the horizontal and vertical axes, in a plane perpendicular respect to the direction of flight. A calibration procedure is applied to the regression outputs to correct residual non-linearities and energy-dependent biases.
Preliminary results on simulated test samples demonstrate good linearity between predicted and true values for both energy and position. After calibration, the relative energy resolution improves with increasing photon energy, reaching the order of 10$\%$ for the highest-energy photons, while position resolutions of 1-3$~mm$ are achieved across most of the detector acceptance. These results highlight the effectiveness of machine learning techniques for offline object reconstruction and calibration in very forward detectors.events in the very forward region.