On the use of neural networks for energy reconstruction in high-granularity calorimeters

Abstract

We contrasted the performance of deep neural networks — Convolutional Neural Network (CNN) and Graph Neural Network (GNN) — to current state of the art energy regression methods in a finely 3D-segmented calorimeter simulated by GEANT4. This comparative benchmark gives us some insight to assess the particular latent signals neural network methods exploit to achieve superior resolution. A CNN trained solely on a pure sample of pions achieved substantial improvement in the energy resolution for both single pions and jets over the conventional approaches. It maintained good performance for electron and photon reconstruction. We also used the Graph Neural Network (GNN) with edge convolution to assess the importance of timing information in the shower development for improved energy reconstruction. We implement a simple simulation based correction to the energy sum derived from the fraction of energy deposited in the electromagnetic shower component. This serves as an approximate dual-readout analogue for our benchmark comparison. Although this study does not include the simulation of detector effects, such as electronic noise, the margin of improvement seems robust enough to suggest these benefits will endure in real-world application. We also find reason to infer that the CNN/GNN methods leverage latent features that concur with our current understanding of the physics of calorimeter measurement.

Dual-Readout Calorimetry for Future Experiments Probing Fundamental Physics

“…Due to sizable uncertainties in the low-energy nuclear physics modeling in GEANT4 [19], these systematic uncertainties cannot be understood until a lepton collider Higgs factory is built and operating (although the CALICE collaboration has done much groundbreaking work using test beams to reduce these systematic uncertainties for single particles in a detector of limited size and the work will certainly benefit from ML techniques such as discussed in Ref. [20]). For high energy jets (order 100 GeV), the individual showers in the jet will overlap even for materials with the smallest Molière radius, leading to the so-called confusion term [14].…”

“…We contrasted [51] the performance of deep neural networks -Convolutional Neural Network (CNN) and Graph Neural Network (GNN) -to current state-of-the-art energy regression methods in a finely 3D-segmented calorimeter simulated by GEANT4. This comparative benchmark gives us some insight about the particular latent signals neural network methods exploit to achieve superior resolution.…”

“…The improved reconstruction performance of the CNN over the simple energy sum, as shown in Figure 25, indicates that the CNN exploits the relationship between the invisible energy and the visible signal in the shower, i.e. the multiplicity and production angle of the secondaries from the hadron interactions (see [51] for details).…”

Precision timing for collider-experiment-based calorimetry

“…More complex reconstruction algorithms can potentially make even better use of the multi-dimensional information provided by highly granular calorimeters with timing. First promising results have been achieved with convolutional and graph neural networks, with a performance increasing with time resolution [19].…”

Perspectives on the Calibration of CNN Energy Reconstruction in Highly Granular Calorimeters

“…We studied the effect of miscalibration and noise on a CNN trained on a Cu/Si module having a 2 × 2 × 4 cm 3 segmentation. In a previous paper, we described the reconstruction performance of this design [2]. The CNN we trained operates as a correction to the sum of energy of the calorimeter cells; the convolutional layers of the CNN consume the raw calorimeter cell energies and we inject the sum of cell energies at a later layer of the network.…”