Heterogeneous Point Set Transformers for Segmentation of Multiple View Particle Detectors

TL;DR

NOvA's data from two sparse views ($XZ$ and $YZ$) require joint prong segmentation and particle-type classification. The authors introduce a heterogeneous point set transformer (HPST) that fuses information across views via heterogeneous attention on a point-cloud representation, achieving substantially lower memory use than CNN baselines. On a large NOvA-derived dataset, HPST reaches a semantic+instance segmentation AUC of $0.968$ while using less than $10\%$ of the memory of traditional CNN methods, and maintains competitive inference speed. This approach offers a scalable, cross-view, memory-efficient solution for multi-view particle detectors at NOvA-scale and beyond.

Abstract

NOvA is a long-baseline neutrino oscillation experiment that detects neutrino particles from the NuMI beam at Fermilab. Before data from this experiment can be used in analyses, raw hits in the detector must be matched to their source particles, and the type of each particle must be identified. This task has commonly been done using a mix of traditional clustering approaches and convolutional neural networks (CNNs). Due to the construction of the detector, the data is presented as two sparse 2D images: an XZ and a YZ view of the detector, rather than a 3D representation. We propose a point set neural network that operates on the sparse matrices with an operation that mixes information from both views. Our model uses less than 10% of the memory required using previous methods while achieving a 96.8% AUC score, a higher score than obtained when both views are processed independently (85.4%).

Figures (3)

• Figure 1: Visual representations of the NOvA detector and data.
• Figure 2: Block diagram of the attention mechanism. The top path describes the intra-view attention mechanism, and the bottom path describes the inter-view mechanism.
• Figure 3: Distribution of prong efficiency and purity