AI-Assisted Object Condensation Clustering for Calorimeter Shower Reconstruction at CLAS12

TL;DR

This work tackles the problem of overabundant fake neutral clusters in CLAS12 ECal-based neutron and photon reconstruction. It introduces an AI clustering approach that fuses GravNet-derived local geometry with a Transformer encoder to create rich hit representations and uses object condensation to form robust clusters in a learned 2D latent space. Trained on one million simulated $e^-+p$ DIS events, the method significantly improves cluster trustworthiness for neutrons from $8.88\%$ to $30.73\%$ and photons from $51.07\%$ to $64.73\%$, illustrating a substantial reduction in misidentified neutral particles. This first application of AI clustering to a hodoscopic detector broadens clustering capabilities in high-energy physics experiments and points to future extensions in multi-object classification and regression tasks, with code available on GitLab.

Abstract

Several nuclear physics studies using the CLAS12 detector rely on the accurate reconstruction of neutrons and photons from its forward angle calorimeter system. These studies often place restrictive cuts when measuring neutral particles due to an overabundance of false clusters created by the existing calorimeter reconstruction software. In this work, we present a new AI approach to clustering CLAS12 calorimeter hits based on the object condensation framework. The model learns a latent representation of the full detector topology using GravNet layers, serving as the positional encoding for an event's calorimeter hits which are processed by a Transformer encoder. This unique structure allows the model to contextualize local and long range information, improving its performance. Evaluated on one million simulated $e+p$ collision events, our method significantly improves cluster trustworthiness: the fraction of reliable neutron clusters, increasing from 8.88\% to 30.73\%, and photon clusters, increasing from 51.07\% to 64.73\%. Our study also marks the first application of AI clustering techniques for hodoscopic detectors, showing potential for usage in many other experiments.

Figures (10)

• Figure 1: Sample $e^-+p$ Monte Carlo event at CLAS12 reconstructed using coatjava plotted in $(\theta,\phi)$ space. Here, $\theta$ and $\phi$ are defined in the lab-frame where the proton target is at rest. Upwards (downwards) facing triangles represent the final state true (reconstructed) particles. Marker size roughly scales with particle energy.
• Figure 2: A schematic of the CLAS12 detector system Burkert_Elouadrhiri_Adhikari_etal._2020. The proton target is positioned within the Central Neutron Detector (CND) and central time-of-flight (CTOF), which are downstream from the backward angle neutron detector (BAND). The electromagnetic calorimeter consists of the PCal, ECin and ECout (the latter two grouped as EC in the figure) subsystems. The forward-angle detector system exhibits six-fold azimuthal symmetry, evident by the distinct corners of its EC, PCal, forward time-of-flight (FTOF), and drift chamber (DC) detectors. The high threshold Cherenkov counter (HTCC) and low threshold Cherenkov counter (LTCC) help discriminate charged particles.
• Figure 3: View of the U, V, W scintillating strip plane layout at CLAS12 Asryan_Chandavar_Chetry_etal._2020.
• Figure 4: Simulated ECal detector response from multiple final state particles in a sample collision event. For each plot, colors represent distinct particles and marker-styles label particle types ($e^-$, $\gamma$, etc). (Left) Scintillator strips are labeled with the truth-level SIDIS particle which caused the hit. (Right) Strips are labeled by the distinct clusters (particles) reconstructed by coatjava . In the bottom right sector, a true generated neutron deposits energy in many strips. When processing this group of hits, the coatjava clustering algorithm reconstructs three neutral particles and misidentifies them as photons. The electron generated in the top-right sector is misidentified as a photon due to a missing drift chamber track match.
• Figure 5: Schematic of the network architecture, highlighting: (a) the higher-level embedding, positional encoding, and feature extraction modules, (b) the sub-networks built into these modules.