Deep Learning for Event Reconstruction in Neutrino Experiments

Neutrino experiments are entering an era of unprecedented detector scale, precision, and data complexity. From liquid argon time projection chambers and water Cherenkov detectors to scintillator-based and hybrid technologies, modern facilities aim to resolve rare, weakly interacting neutrino events with high accuracy while operating in environments dominated by noise, backgrounds, detector effects, or incomplete information. Event reconstruction (transforming raw detector signals into physically meaningful quantities such as interaction vertices, particle identities, energies, directions, and topologies) is therefore central to the success of current and next-generation neutrino physics programs.

Deep learning has rapidly emerged as a transformative tool for this challenge. Convolutional neural networks, graph neural networks, transformers, autoencoders, diffusion models, and physics-informed architectures offer new ways to extract information from sparse, high-dimensional, and irregular detector data. These approaches can improve reconstruction performance, accelerate simulation and analysis workflows, and enable more powerful classification of neutrino interaction channels, including those relevant to oscillation measurements, sterile neutrino searches, supernova neutrino detection, coherent scattering, and beyond-the-Standard-Model physics.

This Research Topic aims to bring together advances at the intersection of neutrino physics, detector instrumentation, and artificial intelligence (AI). We welcome contributions addressing deep-learning-based reconstruction of neutrino events across experimental platforms, including studies of energy and direction reconstruction, particle identification, semantic segmentation, clustering, vertex finding, calorimetry, track-shower separation, timing reconstruction, and background rejection. Submissions may also explore uncertainty quantification, domain adaptation between simulation and data, robustness to detector systematics, explainability, real-time triggering, fast simulation, anomaly detection, and integration of machine learning outputs into physics analyses.

Particular emphasis is placed on methods that connect algorithmic innovation with experimental interpretability and physics performance. Comparative studies, open datasets, benchmark tasks, reproducible workflows, and analyses of failure modes are encouraged, as are interdisciplinary perspectives that bridge high-energy physics, astroparticle physics, computer vision, geometric deep learning, and scientific AI.

By collecting recent developments and future directions, this Research Topic seeks to clarify how deep learning can enhance event reconstruction in neutrino experiments while addressing the practical requirements of reliability, transparency, systematic control, and deployment in large-scale scientific collaborations.