Semantic segmentation with a sparse convolutional neural network for event reconstruction in MicroBooNE

Abstract

We present the performance of a semantic segmentation network, SparseSSNet, that provides pixel-level classification of MicroBooNE data. The MicroBooNE experiment employs a liquid argon time projection chamber for the study of neutrino properties and interactions. SparseSSNet is a submanifold sparse convolutional neural network, which provides the initial machine learning based algorithm utilized in one of MicroBooNE's νe-appearance oscillation analyses. The network is trained to categorize pixels into five classes, which are re-classified into two classes more relevant to the current analysis. The output of SparseSSNet is a key input in further analysis steps.This technique, used for the first time in liquid argon time projection chambers data and is an improvement compared to a previously used convolutional neural network, both in accuracy and computing resource utilization. The accuracy achieved on the test sample is ≥ 99%. For full neutrino interaction simulations, the time for processing one image is ≈ 0.5 sec, the memory usage is at 1 GB level, which allows utilization of most typical CPU worker machine. 42 eSSNet plays an important role in many tasks such as 43 vertex finding, shower reconstruction and energy recon-44 struction, and neutrino selection in the DL-LEE search in 45 MicroBooNE, and is the first usage of SSCN in LArTPC 46 data (beam-off) and realistic MC (neutrino interactions).

Real-time Inference with 2D Convolutional Neural Networks on Field Programmable Gate Arrays for High-rate Particle Imaging Detectors

“…In particular, CNNs have been shown to achieve very high signal selection efficiencies especially when employed in offline physics analyses of LArTPC data. MicroBooNE is leading the development and application of ML techniques, including CNNs, for event reconstruction and physics analysis as an operating LArTPC [27][28][29][30], and CNN-based analyses and ML-based reconstruction are actively being developed for SBN and for DUNE [31,32].…”

Machine Learning in the Search for New Fundamental Physics

“…Moving more toward implementation of deep learning algorithms for reconstruction tasks rather than end-user analysis tasks, Mi-croBooNE has further demonstrated multi-particle type identification [71,213], as well as pixel-level object prediction [214] through a successful first application of a UNet style CNN architecture and semantic segmentation techniques to LArTPC neutrino data. In an effort to minimize computational needs for CNN applications to (particularly sparse) LArTPC data analysis, MicroBooNE has also demonstrated the use of semantic segmentation with a sparse CNN for event reconstruction [215]. This was motivated by Ref.…”

“…SBN's near detector (SBND) has also applied a UResNet network for cosmic background removal [217], demonstrating scalability of larger CNNs for pixel-level signal-background rejection task with larger images [217]. The technique has also been employed by MicroBooNE as part of the experiment's flagship BSM physics search [215].…”

Convolutional Neural Networks for Shower Energy Prediction in Liquid Argon Time Projection Chambers

“…For this study, we focused on two neural network architectures: the residual network [16] and the inception network [17]. Each has found success in previous LArTPC deep learning projects [18,19] The residual network (ResNet [16]), introduced in 2015, was designed to allow deep layers in the network to learn small adjustments to the final output easily. Its central structure is the residual block, a network layer whose output is a sum of its input and a filtered "residual."…”