Particle Hit Clustering and Identification Using Point Set Transformers in Liquid Argon Time Projection Chambers

TL;DR

This work tackles the challenge of reconstructing sparse LArTPC data by replacing dense CNNs with a point-set transformer framework that operates directly on sparse point coordinates across multiple views. It introduces homogeneous and heterogeneous attention, along with grid-based pooling, in a U‑Net–like architecture to perform joint semantic and instance segmentation. The 3D PST and its heterogeneous variant (HPST) outperform graph-attention and CNN baselines in accuracy while delivering substantially lower runtime and memory usage than dense CNN methods, demonstrating a favorable trade-off for large-scale neutrino and dark-matter experiments. The approach enables accurate prong clustering and particle identification with practical performance characteristics, highlighting its potential for scalable LArTPC event reconstruction.

Abstract

Liquid argon time projection chambers are often used in neutrino physics and dark-matter searches because of their high spatial resolution. The images generated by these detectors are extremely sparse, as the energy values detected by most of the detector are equal to 0, meaning that despite their high resolution, most of the detector is unused in a particular interaction. Instead of representing all of the empty detections, the interaction is usually stored as a sparse matrix, a list of detection locations paired with their energy values. Traditional machine learning methods that have been applied to particle reconstruction such as convolutional neural networks (CNNs), however, cannot operate over data stored in this way and therefore must have the matrix fully instantiated as a dense matrix. Operating on dense matrices requires a lot of memory and computation time, in contrast to directly operating on the sparse matrix. We propose a machine learning model using a point set neural network that operates over a sparse matrix, greatly improving both processing speed and accuracy over methods that instantiate the dense matrix, as well as over other methods that operate over sparse matrices. Compared to competing state-of-the-art methods, our method improves classification performance by 14%, segmentation performance by more than 22%, while taking 80% less time and using 66% less memory. Compared to state-of-the-art CNN methods, our method improves classification performance by more than 86%, segmentation performance by more than 71%, while reducing runtime by 91% and reducing memory usage by 61%.

Figures (5)

• Figure 1: Block diagram of the attention mechanism. The top path describes the intra-view attention mechanism, and the bottom path describes the inter-view mechanism. The top section labeled PST is the attention mechanism used in the point set transformer, while HPST uses both the top and bottom sections.
• Figure 2: Architecture of the neural network. The attention block is described in Figure \ref{['fig:attn']}. The number of stages can be arbitrarily increased by adding stages to both the pooling and unpooling sides.
• Figure 3: Distribution of prong efficiency and purity
• Figure 4: Two example events from the test set. The left two columns show the X and Y views of a muon neutrino event with a long muon track (blue), and the right two show an electron neutrino event with a prominent electron shower (black). The top row shows each hit's true particle label and the bottom row shows the network's predicted segmentation each colored according to the particle class that had the majority of hits classified as such in the segment.
• Figure 5: The two example events from figure \ref{['fig:plots']} from the test set visualized as 3D plots.