Search

Machine learning (ML) methodologies, in recent years, have carved a significant niche within the realm of neutrino physics. These tools are not just ancillary but pivotal to the success of emerging experimental endeavors in the field. An overwhelming portion of current neutrino experiments rely heavily on an array of state-of-the-art machine learning tools. These sophisticated ML tools have empowered researchers to derive physical insights of unparalleled depth and clarity, significantly outpacing the potential of traditional data analytics approaches. Central to this achievement are the advanced neural network designs being employed. With the rapid evolution of machine learning techniques across diverse sectors, it is of paramount importance for neutrino physicists to remain updated and adapt these innovations seamlessly into their research workflows. The pace and vigor of advancements in machine learning, especially its applications in neutrino research, are nothing short of remarkable. Given this dynamic landscape, there's a continuous imperative to periodically distill, synthesize, and keep tabs on emerging trends. This paper, in essence, is an endeavor to provide such a synthesized overview, capturing the zeitgeist of ML applications in neutrino physics. [ABSTRACT FROM AUTHOR]

Deep Convolutional Neural Networks (CNNs) have exhibited remarkable eficacy in data analysis across various domains of physical research. Within the realm of particle and high-energy physics, an imperative challenge revolves around the scrutiny of particle tracks acquired through track detectors, exemplified by liquefied argon time projection chambers (LArTPCs). These cutting-edge neural networks have showcased unparalleled performance in tackling this intricate task, demonstrating their profound potential in enhancing particle track analysis. Since the scalability requirement is imposed on neutrino detectors, it is also imposed on CNN model implementations. The scalability for LArTPC data processing is achieved due to the sparsity of the data, which have the form of thin trajectories. To process sparse data, subspecies sparse convolutional networks (SSCNs) and sparse tensor networks (STNs) have been proposed. In this paper, we present the results of semantic segmentation of the LArTPC-simulated PILArNET data using STNs and test various modifications of the classical U-Net architecture: Attention U-ResNet, U-ResNet3+, U-ResNet with an additional deep supervision block, as well as loss functions: focal loss, balanced cross-entropy to improve the accuracy of Michel electron identification. The best results were obtained for balanced cross-entropy. Further improvement is possible by combining several methods. [ABSTRACT FROM AUTHOR]