Bulletin of the American Physical Society
APS April Meeting 2022
Volume 67, Number 6
Saturday–Tuesday, April 9–12, 2022; New York
Session W09: Data Analysis, AI and ML IIIRecordings Available
|
Hide Abstracts |
Sponsoring Units: DPF GDS Chair: Zoya Vallari, Caltech Room: Salon 3 |
Monday, April 11, 2022 5:45PM - 5:57PM |
W09.00001: Application of machine learning to find anomalous events in the LZ data Maris Arthurs LUX-ZEPLIN (LZ) is a dark matter direct detection experiment using a dual-phase xenon time projection chamber with a 7-ton active volume, expecting science results in 2022. Anomalies are expected in data in the early stages of the experiment, such as from misclassification of pulses and interaction types, as well as detector pathologies. Dimensional reduction is an unsupervised Machine Learning technique that can effectively identify anomalous events. High-dimensional data can be mapped with minimal loss of structure to a low-dimensional space, where the data can be visualized and clustered. In this presentation, we will discuss the application of the uniform manifold approximation and projection dimensional reduction algorithm to reduce the early LZ data to a 2D space for the identification of similar populations and anomalous events. |
Monday, April 11, 2022 5:57PM - 6:09PM |
W09.00002: Classifying scintillation and ionization signals in xenon detectors with machine learning Sophia Farrell Liquid-xenon time projection chambers (TPCs) have set leading limits on WIMP dark matter. These detectors also can probe a number of electrophilic and nucleophilic processes beyond the Standard Model. The XENONnT experiment will be sensitive to new low-threshold signals, such as the neutrino magnetic moment and solar axions. These searches require ultra-low energy thresholds and backgrounds. Interactions in xenon TPCs are characterized by two signals: prompt scintillation and delayed ionization. At high energies, these signals are trivial to classify. However, at threshold-scale energies, classifying scintillation and ionization signals is nontrivial. Misclassification can result in increased accidental coincidence background rates or reduced signal detection efficiency. We present a machine learning approach for low-energy signal classification in XENONnT called a Naive Bayes classifier. Compared with standard techniques, our method infers signal probabilities, allowing experiments to improve upon current deterministic decision boundaries for signal classification and improve detection efficiency. |
Monday, April 11, 2022 6:09PM - 6:21PM |
W09.00003: A Novel Method Using Generative Neural Networks for Event Reconstruction in Water Cherenkov Detector Mo Jia Large water Cherenkov detectors are widely used in the detection of neutrinos and nucleon decays. The current approaches to reconstruct events in these detectors involve maximum-likelihood methods in which a likelihood function incorporated with the observed signals on the photosensors is constructed. The likelihood is maximized under different configurations of the event hypothesis to find the best fit for each event. Here we introduce a new way which is based on the traditional likelihood but at the same time employs generative neural networks in place of many templates of traditionally simulated events. The networks use inputs that characterize an event in the detector and predict probability density distributions for the signals at each photosensor. The networks are trained with Monte Carlo samples of electrons and muons. We present the initial results from this deep-learning based approach including demonstrations of particle identification and energy reconstruction. |
Monday, April 11, 2022 6:21PM - 6:33PM |
W09.00004: Sparse Convolutional Neural Networks for NOνA Haejun Oh The NOνA (NuMI Off-axis ν Appearance) experiment measures neutrino oscillations in a nearly pure muon neutrino beam over a 810 km baseline. NOνA has successfully used a Convolutional Neural Network (CNN) as its main event selector and particle identifier for oscillation analysis since 2016. Despite having over 90% efficiency in classifying all neutrino types, our current CNN requires significant GPU resource during training and can only be applied to a window of activity around the beginning of each event. In order to reduce the computational cost, we have implemented a Sparse Convolutional Neural Network (SCNN) which performs convolutions only when the center pixel of the receptive field is non-zero. This reduces the size of training data and improves throughput. First, we have implemented a sparse MobileNet architecture using the Minkowski Engine package and achieved 89% accuracy with 83% decrease in GPU memory used. In addition, we have implemented a sparse FishNet architecture which resulted in training accuracy of ~90%. In the future, we hope to also incorporate semantic and instance segmentation simultaneously into the FishNet architecture for full end-to-end reconstruction. |
Monday, April 11, 2022 6:33PM - 6:45PM |
W09.00005: Reducing sensitivity to systematic uncertainties of the deep neural networks employed in the NOvA experiment Kevin Mulder In the NOvA experiment, with its pixelated detectors, measured events can be recorded in an image format. This allows for the usage of powerful image identification techniques such as convolutional neural networks (CNN’s) for the purposes of event classification. |
Monday, April 11, 2022 6:45PM - 6:57PM |
W09.00006: The Noble Element Simulation Technique (NEST): Recent Updates and Improvements Matthew M Szydagis Noble element detectors (two-phase emission detectors, liquid phase-only detectors, etc.) have many applications in modern research. For example, they are broadly used in dark matter registration, non-standard neutrino interactions searches and even Standard Model processes observation (for example, coherent elastic neutrino-nucleus scattering (CEvNS) studies). Modeling signal generation from these complicated interactions requires precise simulations. The main problem of modeling such phenomena is that various theoretical predictions are inconsistent with each other and compared to experimental data. The Noble Element Simulation Technique (NEST) provides a semi-empirical solution for modeling xenon and argon detector response by combining theoretical models (such as Lindhard and its variations) and actual experimental data. NEST can simulate not only the median scintillation and ionization yields for various interaction types, fields (including zero field) and energies (from sub-keV to tens of MeV), but also detector-specific response (electron extraction efficiency, basic waveforms, electron drift speed in liquid/gas phase, etc.). Currently, NEST exists in three forms: as a GEANT4 library, a separate C++ package, and a standalone Python package (nestpy). At this talk, recent NEST updates will be discussed and future plans will be presented. |
Monday, April 11, 2022 6:57PM - 7:09PM |
W09.00007: Panoptic Segmentation for Particle Identification in LArTPC Detectors Carlos E Sarasty The Deep Underground Neutrino Experiment (DUNE) will use the liquid argon time projection chamber (LArTPC) technology for its near and far detectors. LArTPC detectors can collect high-resolution data of charged particles’ trajectories. An example of this type of detector is ProtoDUNE-SP that is the prototype of the single-phase DUNE far detector using full-scale components and a charged-particle beam that allows measuring the detector’s calorimetric response to hadronic particles and electromagnetic showers. Convolutional Neural Networks have been developed and employed in the analysis of scientific data from the ProtoDUNE detector, which exploit the advantages of a liquid argon time projection chamber (LArTPC). Despite the high-resolution images and the fine details that the detector can capture, the classification of the different types of particles and interactions is still a challenge. With this motivation, we present the details of a multi-task reconstruction algorithm using a Sparse Convolutional Neural Network capable of panoptic segmentation – that is simultaneously generating a pixel-by-pixel particle ID and clustering pixels into objects. |
Monday, April 11, 2022 7:09PM - 7:21PM |
W09.00008: Deep-Learning-Based Kinematic Reconstruction for DUNE Wenjie Wu The Deep Underground Neutrino Experiment (DUNE) is a next-generation long-baseline neutrino oscillation experiment based on liquid argon TPC (LArTPC) technology. While LArTPC technology provides excellent spatial resolution, high neutrino detection efficiency, and superb background rejection, it poses significant reconstruction challenges. Deep learning methods, in particular Convolutional Neural Networks (CNNs), have been successfully used in classification problems such as particle identification in DUNE and other neutrino experiments. However, deep learning methods for regression problems, such as the reconstruction of neutrino energies and final state particle momenta, have not yet been developed. Here we design, train, and test two CNN-based methods, using 2-D and 3-D data, for the reconstruction of final state particle direction and energy, as well as neutrino energy. Combining our models with particle masses yields a fully AI-based reconstruction chain producing the four-vector momenta of the final state particles. Compared to a traditional method, our models show considerable performance improvements for both νe and νμ scenarios. |
Monday, April 11, 2022 7:21PM - 7:33PM |
W09.00009: Anticipating antineutrinos at the Theia detector Zara Bagdasarian, Stephane Zsoldos, Stephen Dye, Gabriel D Orebi Gann Theia is a proposed large-scale novel neutrino detector designed to discriminate between Cherenkov and scintillation signals and enable a broad physics program. The baseline design consists of a tank filled with water-based liquid scintillator (WbLS), a novel target which would combine reconstruction of particle direction from the Cherenkov signal, with the excellent energy resolution and low threshold of a scintillator detector. This talk will focus on the sensitivity towards the detection of low-energy antineutrinos in the context of the 25-ktonne Theia detector at Sanford Underground Research Facility (SURF). Currently, only two detectors in the world have measured geoneutrinos, antineutrinos originating from the radioactive decays of long-lived isotopes naturally present in the crust and mantle. It is critical to gather geoneutrino data from more detectors at various locations around the world as it can vary drastically. We demonstrate Theia's sensitivity to measure the antineutrino fluxes via Inverse-Beta Decay (IBD) interactions and collect an unprecedented amount of geoneutrino data in only one year to extract the crust/mantle contribution. |
Follow Us |
Engage
Become an APS Member |
My APS
Renew Membership |
Information for |
About APSThe American Physical Society (APS) is a non-profit membership organization working to advance the knowledge of physics. |
© 2024 American Physical Society
| All rights reserved | Terms of Use
| Contact Us
Headquarters
1 Physics Ellipse, College Park, MD 20740-3844
(301) 209-3200
Editorial Office
100 Motor Pkwy, Suite 110, Hauppauge, NY 11788
(631) 591-4000
Office of Public Affairs
529 14th St NW, Suite 1050, Washington, D.C. 20045-2001
(202) 662-8700