Bulletin of the American Physical Society
6th Joint Meeting of the APS Division of Nuclear Physics and the Physical Society of Japan
Sunday–Friday, November 26–December 1 2023; Hawaii, the Big Island
Session L08: Minisymposium: Applications of Advanced Statistics and Machine Learning Methods in Nuclear Physics III |
Hide Abstracts |
Chair: Amy Lovell, Los Alamos Natl Lab Room: Hilton Waikoloa Village Kohala 1 |
Friday, December 1, 2023 9:00AM - 9:15AM |
L08.00001: Towards Automation for γ-Ray Spectroscopy Tamas A Budner, David Lenz, Michael P Carpenter, Sven Leyffer, Filip G Kondev, Amel Korichi, Torben Lauritsen, Thomas F Lynn, Marco Siciliano Over the past century, γ-ray spectroscopy has been a powerful experimental tool for probing the structure of atomic nuclei. Incremental improvements in radiation detectors and ion accelerator technologies have dramatically increased both the quality and quantity of the nuclear data that can be collected in a single measurement. However, traditional methods of analyzing spectroscopic data towards the goal of constructing accurate nuclear decay schemes have remained largely unchanged over time. Visually inspecting one- and two-dimensional histograms, time-gating on γ-ray coincidence data, fitting spectra, and building upon previously reported level diagrams within the academic literature are time-consuming and error-prone processes, which would likely benefit from the application of modern data science techniques. Here, we discuss the development of computational tools for analyzing high-statistics, γ-ray datasets, presenting preliminary capabilities benchmarked against evaluated nuclear data. In addition to automating familiar analysis steps, such as multidimensional background subtraction and Gaussian peak-fitting, we also propose a reformulation of the scheme-building procedure as a Bayesian inverse problem. Using existing numerical optimization methods, this novel approach to spectroscopic analysis enables the recovery of a directed level-scheme graph from symmetric γ-γ coincidence matrices. |
Friday, December 1, 2023 9:15AM - 9:30AM |
L08.00002: Machine learning based design optimization for the search of neutrinoless double-beta decay with LEGEND Ann-Kathrin Schuetz The in-situ production of long-lived radio-isotopes by cosmic muon interactions may generate a non-negligible background for rare event searches deep underground. The delayed decay of 77(m)Ge has been identified as the dominant in-situ cosmogenic contributor for a neutrinoless double-beta decay search with 76Ge. The future ton-scale LEGEND-1000 experiment requires a total background of < 10-5 cts/(keV·kg·yr). Neutron backgrounds have a strong dependence on laboratory depth, shielding material, and cryostat design. The addition of passive neutron moderators results in a reduced background contribution. Therefore, Monte Carlo studies using a custom simulation module based on Geant4 are performed to optimize the moderator screening effect. However, using traditional Monte Carlo simulations a full optimization of a many parameter space may still be a time consuming and difficult task to address. Machine learning can help in both speeding up common modeling problems, as well as help to minimize the application of computational expensive standard Monte Carlo methods. The Multi-Fidelity Gaussian Process based study presented in this talk aims to demonstrate a techniques on a small-scale application, which then is gradually adaptable to the more ambitious task of exploring innovative solutions to the design of detectors for future 76Ge experiments. |
Friday, December 1, 2023 9:30AM - 9:45AM |
L08.00003: Using Convolutional Neural Networks to Classify Scintillator Data Adam Hartley, Sean N Liddick, Geir Ulvik, Morten Hjorth-Jensen, Aaron Chester Isomeric states are sensitive to changes in nuclear structure. Being able to observe evidence of isomeric transitions is important to expanding our knowledge of nuclear structure as a whole. A monolithic inorganic planar scintillator coupled to a position-sensitive photomultiplier tube readout is used to record the energy deposition in the detector through the light emission due to β particles, internal conversion electrons, and γ rays. The positions and energies of which are associated with previously identified nuclei to allow introspection into changes in the structure of the nucleus. Of particular interest is the detection of two interactions within the scintillator that occur separated in time and/or space, which are currently unrecoverable by the current analysis pipeline. This drives the development of a method to produce artificial experimental data to be used as training data for convolutional neural networks. This method has promising results in classifying single and multiple interactions in artificial scintillator data. |
Friday, December 1, 2023 9:45AM - 10:00AM |
L08.00004: CARIBU-matic and the MUSIC ML project: examples of machine-learning applications for beam tuning and experimental data analysis/classification Daniel Santiago-Gonzalez, Melina Avila, Prasanna Balaprakash, Heshani Jayatissa, Krishnan Raghavan, Nathan Callahan During experiments at accelerator facilities, timely production and transport of the requested beam to the users target station is crucial for both the experiment success and to maintain high user satisfaction. During the data collection phase of an experiment, researchers need fast and accurate methods to process and classify the data produced by their detector systems to verify if changes to the experimental conditions need to be carried out. |
Friday, December 1, 2023 10:00AM - 10:15AM |
L08.00005: Offline reinforcement learning for closed-loop control of the VENUS ion source Yue Shi Lai While deep reinforcement learning (RL) for control offers the promise of optimal steering and improved resource utilization for electron cyclotron resonance (ECR) ion sources, there are risks associated with opening a critical user facility equipment with complex phase space to the free exploration of online RL. At the same time, accurate modeling of an ECR ion source remained challenging. We discuss the development of an automated, closed-loop control system using RL for VENUS (Versatile ECR ion source for NUclear Science), the third-generation superconducting electron cyclotron resonance ion source for the 88-Inch Cyclotron at the Lawrence Berkeley National Laboratory. Risks are minimized by utilizing offline RL on historical recordings of human operators interacting with the ion source without involving any simulation or surrogate model. We discuss our first result using behavior cloning (BC) and conservative Q-learning (CQL). We further provide lessons and an outlook on offline RL for secure, AI-based closed-loop control at DOE user facilities. |
Friday, December 1, 2023 10:15AM - 10:30AM |
L08.00006: Atomic Masses with Machine Learning for the Astrophysical R-process Mengke Li, Trevor M Sprouse, Bradley S Meyer, Matthew R Mumpower The astrophysical r process plays a vital role in the production of heavy elements. Modeling of the r process is sensitive to masses and further requires knowledge of masses beyond current experimental reach. Therefore, simulations of the r process offer a unique test bed for predicting mass extrapolations. We take a Machine-Learning (ML) approach to model the masses across the entire chart of nuclides. For the first time, we simulate r-process nucleosynthesis with an ML mass model. We compare simulated abundances to solar data in order to evaluate the model's performance far from stability. The resulting r-process abundances up to thorium and uranium qualitatively match those of the observed solar system abundance pattern, with the characteristic peaks well positioned. We propagate the mass uncertainties obtained from the ML model to r-process abundance yields to estimate an uncertainty band associated with our approach. The size of the uncertainty band is approximately one order of magnitude which aligns with the uncertainty reported in existing literature. |
Friday, December 1, 2023 10:30AM - 10:45AM |
L08.00007: Machine learning assisted filtering approach for ion source optimization and control victor watson, Heather L Crawford, Marco Salathe, Damon Todd The superconducting electron cyclotron resonance (ECR) ion source VENUS is the primary injector for the 88" Inch Cyclotron at LBNL and a copy of it is being installed as the primary driver for the FRIB linac. Optimization of these sources is made difficult by vast operation control spaces, a lack of useful models, and responses to control parameters that can be highly nonlinear. To guide source operation, we are employing Machine Learning tools and over a year of collected control and diagnostic data to produce local approximation functions that can be used to direct operators to control parameter changes that will maximize beam current while maintaining stability. We will describe a filtering approach that takes into account environmental conditions as well as the past and present state of the source to predict what change in the input has the highest probability of increasing the beam current and help guide the operator in the decision making process. |
Friday, December 1, 2023 10:45AM - 11:00AM |
L08.00008: Quantifying uncertainty of nuclear properties within machine learning frameworks Mengyao Huang, Kyle A Wendt, Nicolas F Schunck, Xiao Chen In this work, we use deep learning technique to learn properties of atomic nuclei while accounting for the underlying neural network uncertainty that arises from the complex parameter manifold. While ensemble uncertainties offer a first insight into the confidence of machine learning predictions, they are computationally intractable in all but the simplest problems. We instead apply an approximation of the ensemble uncertainty that can be computed both in simple problems as well as in complex machine learning architectures. In this demonstrative calculation, we use nuclear masses computed by density functional theory (DFT). We train on a subset of the computed mass table and examine the quality of the predicted masses, both in interpolative and extrapolative regimes, as well as the resulting ensemble uncertainty within this approximation scheme. |
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