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 C08: Minisymposium: Applications of Advanced Statistics and Machine Learning Methods in Nuclear Physics I |
Hide Abstracts |
Chair: Michelle Kuchera, Davidson College Room: Hilton Waikoloa Village Kohala 1 |
Tuesday, November 28, 2023 7:00PM - 7:30PM |
C08.00001: Normalizing flows for uncertainty quantification in a Bayesian framework Invited Speaker: Yukari Yamauchi Uncertainty quantification of theoretical models results in a posterior distribution of credible model parameters, and acquiring accurate statistics from such distributions is an essential building block for a model prediction of observations in future experiments. Posterior distributions can be continuously updated when new experimental data or theoretical model improvements become available, requiring new samplings of the model parameters at each update. Standard Monte Carlo methods, however, are typically inefficient for the high-dimensional distributions encountered in many theoretical models. Normalizing flow serves as a powerful tool to accelerate the sampling from these frequently-changing posterior distributions. A normalizing flow is a map that induces a non-trivial distribution function, such as posterior distributions, from the Gaussian distribution of the same dimension. Once such a normalizing flow is obtained, sampling of the model parameters can be done from the Gaussian distribution, followed by the application of normalizing flow. This sampling process can be done in parallel without any autocorrelations between samples, thus achieving a dramatic speed-up compared to the standard Monte Carlo methods. In this talk, we will discuss a method for training and updating normalizing flows for posteriors distributions via machine learning and demonstrate the method within covariant density functional theory. |
Tuesday, November 28, 2023 7:30PM - 7:45PM |
C08.00002: Neural-network quantum states for ultra-cold Fermi gases Jane M Kim, Bryce Fore, Gabriel M Pescia, Jannes Nys, Giuseppe Carleo, Alessandro Lovato, Stefano Gandolfi, Morten Hjorth-Jensen Ultra-cold Fermi gases display diverse quantum mechanical properties, including the transition from a fermionic superfluid BCS state to a bosonic superfluid BEC state, which can be probed experimentally with high precision. However, the theoretical description of these properties is challenging due to the onset of strong pairing correlations and the non-perturbative nature of the interaction among the constituent particles. This work introduces a novel Pfaffian-Jastrow neural-network quantum state that includes backflow transformation based on message-passing architecture to efficiently encode pairing correlations. Our approach offers substantial improvements over comparable ansätze constructed within the Slater-Jastrow framework and out-performs state-of-the-art diffusion Monte Carlo methods. We observe the emergence of strong pairing correlations through the opposite-spin pair distribution functions, and we compute the pairing gap. Moreover, we demonstrate that transfer learning stabilizes and accelerates the training of the neural-network wave function, enabling the exploration of the BCS-BEC crossover region near unitarity. Our findings suggest that neural-network quantum states provide a promising strategy for studying ultra-cold Fermi gases. |
Tuesday, November 28, 2023 7:45PM - 8:00PM |
C08.00003: Optimizing Fermionic Neural Networks with Decision Geometry Mehdi Drissi, James W Keeble, Arnau Rios Recently, Variational Monte Carlo (VMC) solutions to the quantum many-body problem have experienced tremendous progress thanks to the use of neural network quantum states [1, 2, 3]. While more and more sophisticated ansätze have been designed to tackle a wide variety of many-body problems, little progress has been made on their optimization process. In this talk, we will revisit the Kronecker Factored Approximate Curvature (KFAC), one of the main optimizers used for the most challenging many-body systems [2, 4]. After exposing how KFAC is fundamentally unfit for VMC with Fermionic Neural Networks (FNNs), I will discuss the design of a novel optimization strategy based on decision geometry [5]. As a test bench, we consider a FNN modelling polarized fermions interacting in an harmonic trap. Preliminary results will be reported, showing how this new optimizer outperforms KFAC in terms of stability, accuracy and speed of convergence. Beyond VMC, the versatility of this approach suggests that decision geometry could provide a solid foundation for accelerating a broad class of machine learning problems.
[1] G. Carleo and M. Troyer. Science, 355(6325):602–606, 2017.
[2] D. Pfau, J. S. Spencer, A. G. Matthews, and W. M. C. Foulkes. Phys. Rev. Res., 2(3):033429, 2020. [3] A. Lovato, C. Adams, G. Carleo, and N. Rocco. Phys. Rev. Res., 4(4):043178, 2022. [4] J. Martens and R. Grosse. In International conference on machine learning, pages 2408–2417. PMLR, 2015. [5] A. P. Dawid. Ann. Inst. Stat. Math., 59:77–93, 2007. |
Tuesday, November 28, 2023 8:00PM - 8:15PM |
C08.00004: Testing neural networks for classifying multi-neutron decay measurements of neutron-unbound systems Thomas Redpath, Jaylen I Rasberry, Clifton D Kpadehyea The MoNA Collaboration investigates neutron-unbound systems using a set of large-area, high-efficiency neutron detectors, the Modular Neutron Array (MoNA) and the Large multi-Institutional Scintillator Array (LISA). Together with the Sweeper magnet and its ancillary detectors, MoNA-LISA enables invariant mass spectroscopy experiments to study neutron-unbound nuclei around and beyond the dripline thus providing information to benchmark nuclear structure models. A crucial step in the analysis of systems that decay by emitting multiple neutrons involves classifying events according to the number of neutrons detected. To address this, small neural networks (< 100 nodes) are being tested as a means to improve the efficiency of the classification process. Preliminary results will be presented from tests with data from two previous MoNA-LISA experiments to measure two-neutron-unbound systems 26O and 10He. |
Tuesday, November 28, 2023 8:15PM - 8:30PM |
C08.00005: Matrix Model Emulators Danny Jammooa, Patrick Cook Presenting a novel approach to a pervasive issue in quantum physics: the determination of extremal eigenvalues of an excessively large Hamiltonian matrix. Existing techniques, ranging from Monte Carlo simulations to variational methods, often fail when a control parameter in the Hamiltonian matrix reach beyond certain limits. The focus of this study is on a new class of implicit deep learning algorithms, termed Matrix Model Emulators (MMEs). Emulators have the potential of bypassing the computational burdens of complex scientific calculations without compromising accuracy. MMEs is a form of Machine Learning that originally was designed to leverage the structure of eigenvector continuation, enabling the calculation of extremal eigenvalues and other observables without the explicit formation of norm and Hamiltonian matrices. However, MME can be seen as a new implicit deep learning architecture. Showcasing a subset of MMEs known as Matrix Eigenvalue Emulators (MEEs) on an array of nuclear physics Hamiltonians and MMEs on a selection of challenging interpolation problems. Furthermore, reveal MMEs/MEEs effectiveness in identifying branch points in the complex plane. The results offer valuable insights for further development and potential applications of MMEs in nuclear physics and beyond. |
Tuesday, November 28, 2023 8:30PM - 8:45PM |
C08.00006: Novel emulators for fission event generators Amy E Lovell, Arvind T Mohan Fission event generators, such as the LANL-developed code CGMF, are being used with increasing frequency in the analysis and interpretation of experimental measurements. These calculations begin with information on fission fragment initial conditions in mass, charge, kinetic energy, spin, and parity, and follow the de-excitation of the fission fragments through prompt neutron and γ-ray emission, conserving energy, momentum, and angular momentum throughout the process. For each fission event, information about the fragment initial conditions are recorded, along with the multiplicity of the emitted neutrons and γ rays, and their energies and directions. The breadth of the observables that are calculated along with the multiple physics models that are included in the complete calculation require the development of a novel emulator to speed up this type of calculation. Here we discuss the construction of an emulator for CGMF that takes as input the fission fragment initial conditions and uses a combination of classifiers and probabilistic mixture density networks to emulate the full decay. The speedup gained from this type of emulator opens the door to consistent optimization of all prompt fission observables. |
Tuesday, November 28, 2023 8:45PM - 9:00PM |
C08.00007: Large Scale Computation of Nuclear Ground States with Machine Learning and Supercomputing Corey Adams Solving many-body quantum systems for their ground state energy is a computationally complex challenge with a rich history. In this talk, we present the latest developments in a scalable algorithm using machine learning surrogate models for nuclear wavefunctions, which scales to leadership class supercomputers and large nuclei. The focus of this talk will be on the computational techniques and machine learning aspects of the algorithm, as well as the path towards nuclei of experimental interest (such as Germanium and up to Xenon) on exascale super computers. |
Tuesday, November 28, 2023 9:00PM - 9:15PM |
C08.00008: Application of machine learning techniques for the ALICE TPC space-charge distortion correction and for particle tracking in Si detectors Hitoshi Baba A Large Ion Collider Experiment (ALICE) is an experiment at the Large Hadron Collider (LHC) which aims to understand the most basic properties of Quantum Chromodynamics (QCD) by observing the quark-gluon plasma (QGP) created in relativistic heavy-ion collisions. The ALICE detector has been largely upgraded during the LHC Long Shutdown to become capable of collecting Pb-Pb collision data at an unprecedented interaction rate of 50 kHz. The Time Projection Chamber (TPC) is the main tracking detector of ALICE. Distortions of the electron drift paths caused by ion backflow from the readout chambers significantly affect the TPC measurements and therefore must be corrected in order to reach the intrinsic detector resolution. The most challenging aspect of the correction is posed by the calibration of distortion fluctuations relevant on time scales in the order of 10 ms. A framework for the distortion fluctuation correction using machine learning techniques is under development and its current status will be discussed. |
Tuesday, November 28, 2023 9:15PM - 9:30PM |
C08.00009: Utilizing machine learning for fast-timing calibration between LaBr3(Ce) detectors in the neutron-rich N = 20 and N=50 regions Tawfik M Gaballah, Benjamin P Crider, Sean N Liddick, Sean P Burcher, M.P. Carpenter, James J Carroll, Aaron Chester, Christopher J Chiara, Kathrine Childers, Partha Chowdhury, Patrick A Copp, Jason T Harke, Daniel E Hoff, Kay Kolos, Edward Lamere, Rebecca Lewis, Brenden R Longfellow, Stephanie Lyons, Mejdi J Mogannam, Shree K Neupane, Timilehin H Ogunbeku, David Perez-Loureiro, Christopher J Prokop, Daniel M Rhodes, Andrea Richard, Olalekan A Shehu, Durga P Siwakoti, Dylan C Smith, Mallory K Smith, Aaron S Tamashiro, Ronald Unz, Yongchi Xiao The nuclear shell evolution can be attributed to changing proton and neutron numbers within the nucleus. Nuclear transition rates, which significantly depend on a precise measurement of level lifetimes, are sensitive indicators for investigating the nuclear shell evolution. Experiments using β decay in the neutron-rich N=20 and N=50 regions were performed at the National Superconducting Cyclotron Laboratory (NSCL). β decays were correlated with the implantation of radioactive nuclei, using a CeBr3 scintillator coupled to a Position-Sensitive Photomultiplier Tube (PSMPT), through spatial and temporal analysis techniques. In these experiments, 15 LaBr3(Ce) detectors were employed for γ radiation detection and fast timing measurement. Time-difference spectra between β decays and γ radiation detection were used to measure half-lives. Calibration of the LaBr3 timing response relative to the CeBr3 was performed along with further corrections for the energy dependent time-walk effects were made using machine learning techniques in the neutron-rich N=20 and N=50 regions. Validation results from comparing the output from the machine learning model with the output from the analytical technique used in previous analysis will be presented. |
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