Bulletin of the American Physical Society
APS April Meeting 2023
Volume 68, Number 6
Minneapolis, Minnesota (Apr 15-18)
Virtual (Apr 24-26); Time Zone: Central Time
Session QQ03: V: Machine learning and AI |
Hide Abstracts |
Sponsoring Units: DPF Chair: Dipangkar Dutta, Mississippi State University Room: Virtual Room 3 |
Tuesday, April 25, 2023 12:00PM - 12:12PM |
QQ03.00001: Deep Learning Symmetries and Lie Groups Roy T Forestano, Konstantin T Matchev, Katia Matcheva, Alexander Roman, Sarunas Verner, Eyup Bedirhan Unlu We design a deep-learning algorithm for the discovery and identification of the continuous group of symmetries present in a labeled dataset. We use fully connected neural networks to model the symmetry transformations and the corresponding generators. We construct loss functions that ensure that the applied transformations are symmetries and that the corresponding set of generators forms a closed (sub)algebra. Our procedure is validated with several examples illustrating different types of conserved quantities preserved by a symmetry. In the process of deriving the full set of symmetries, we analyze the complete subgroup structure of the rotation groups SO(2), SO(3), and SO(4) and of the Lorentz group SO(1,3). Other examples include SO(10), squeeze mapping, and piece-wise discontinuous labels, demonstrating that our method is completely general, with many possible data science applications. Our study also opens the door for using a machine learning approach in the mathematical study of Lie groups and their properties. |
Tuesday, April 25, 2023 12:12PM - 12:24PM |
QQ03.00002: Using XGBoost to correctly pair muons in events with two dimuons from off-shell boson decays Stephen D Butalla, Spencer Hirsch, Marcus Hohlmann Machine learning applications for tasks in high energy physics have proliferated in recent years—applications can be found in the realm of jet and particle classification, event selection, and triggering to name a few. One area that has received scant attention is the reconstruction of decay products from off-shell parent particles. For instance, in a physics analysis where the final state contains two pairs of muons (known as dimuons), one can reconstruct the final-state muon pairs by calculating the invariant masses of the dimuon pairs and then finding the correct combination by minimizing the difference in the invariant masses of the dimuon pairs if the parent bosons of the dimuons are on their mass shells. This method becomes invalid when these muons decay from off-shell bosons, necessitating a different method to determine the correct muon pairing. To tackle this problem, we present an eXtreme Gradient Boosted (XGBoost) decision tree model trained on Monte-Carlo simulated data that classifies the correct and incorrect pairings of the muons in the final-state dimuons. We also present our results of using data augmentation, feature engineering, and hyperparameter tuning to maximize the performance metrics of our model. Preliminary results indicate a maximum accuracy of 0.972, a maximum area-under-the-curve of 0.997, and a maximum Matthews correlation coefficient of 0.936. The aim of these studies is to eventually create an extensible model which can be employed in searches for bosons in dark matter models. |
Tuesday, April 25, 2023 12:24PM - 12:36PM |
QQ03.00003: Trigger-Level Event Reconstruction for Neutrino Telescopes Using Sparse Submanifold Convolutional Neural Networks Felix Yu, Jeffrey P Lazar, Carlos A Arguelles Convolutional neural networks (CNNs) have seen extensive applications in scientific data analysis, including in neutrino telescope experiments. However, the data from these experiments present numerous challenges to CNNs, such as non-regular geometry, sparsity, and high dimensionality. As a result, CNNs are highly inefficient on neutrino telescope data, and require significant pre-processing that results in information loss. We propose utilizing sparse submanifold convolutions as a solution to these issues. We aim to show that CNNs using these sparse submanifold convolutions achieves the competitive performance expected from a machine learning algorithm, while running orders of magnitude faster on both GPU and CPU compared to a traditional CNN. As a result of this speedup, these networks are capable of handling the trigger-level event rate for experiments such as IceCube. |
Tuesday, April 25, 2023 12:36PM - 12:48PM |
QQ03.00004: Graph Neural Network-Based Optimization of IceCube-Gen2 Geometry Tong Zhu, Miaochen Jin, Carlos A Arguelles Delgado IceCube-Gen2 is the planned upgrade of the IceCube Neutrino Observatory at the South Pole designed to probe the high-energy neutrino sky from TeV to EeV energies, with a ten times more volume than the current IceCube detector. As more strings will be included at a larger separation distance, we need a geometry that provides us with an optimal ability in capturing the events and recording their information despite the increased spacing between strings. In this study, we utilize graph neural networks as the reconstruction method for IceCube-Gen2, and evaluate the performance of this algorithm under 16 proposed candidate geometries, arranging a prototype detector with 196 strings in 4 different shapes and 4 different geometric areas. |
Tuesday, April 25, 2023 12:48PM - 1:00PM |
QQ03.00005: A particle tracker for diffusion cloud chambers Murtuza S Taqi, Timothy D Wiser, Zoe Rechav
|
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