Core Concepts
NuGraph2 is a powerful Graph Neural Network designed for low-level reconstruction of simulated neutrino interactions in a Liquid Argon Time Projection Chamber detector, achieving high efficiency in background filtering and semantic labeling.
Abstract
I. Introduction:
Liquid Argon Time Projection Chamber (LArTPC) technology provides detailed information on particle interactions.
NuGraph2 utilizes a Graph Neural Network for reconstructing neutrino interactions in LArTPC detectors.
II. Experimental Setup:
Utilizes an open dataset from the MicroBooNE collaboration for training and testing.
Constructs heterogeneous graph objects representing each neutrino interaction.
III. Model Architecture:
Consists of an encoder, message-passing engine, and decoders for filter and semantic outputs.
Utilizes categorical embedding and cross-attention mechanisms for efficient convolution operations.
IV. Training:
PyTorch with PyG used for model training with AdamW optimizer and OneCycleLR scheduler.
V. Results:
Achieves high recall and precision in filter prediction and semantic labeling tasks.
VI. Summary:
NuGraph2 offers a versatile solution for particle reconstruction in neutrino physics, adaptable to various detector technologies.
Stats
Liquid Argon Time Projection Chamber (LArTPC) detector technology offers high-resolution information on particle interactions.
The network identifies primary physics interactions with 98.0% efficiency and labels particles with 94.9% efficiency.