Bulletin of the American Physical Society
APS April Meeting 2024
Wednesday–Saturday, April 3–6, 2024; Sacramento & Virtual
Session M17: Data Science and AI/ML in Physics |
Hide Abstracts |
Sponsoring Units: GDS Chair: Kevin Pedro, Fermi National Accelerator Laboratory Room: SAFE Credit Union Convention Center Ballroom B4, Floor 2 |
Friday, April 5, 2024 1:30PM - 1:42PM |
M17.00001: Adaptive machine learning with hard physics constraints for 6D phase space diagnostics of intense charged particle beams Alexander Scheinker Advanced particle accelerators, such as the FACET-II plasma wakefield acceleration test facility are creating intense high energy (10s of GeV) high charge electron beams. For example, a signle electron bunch in FACET-II can have up to 2 nC of charge within an incredibly short bunch length (σt) of 1-100 femtoseconds (1 fs = 10-15 s) resulting in peak currents of up to 200 kA. Such highly energetic, intense, and short beams are difficult to image because they may damage intercepting diagnostics such as scintillating screens and because their lengths are exceeding the resolution of deflecting cavity-based diagnostics, such as TCAVs whose resolution is ~3 fs. Machine learning methods have shown promise towards enabling virtual diagnostics of such beams, but so far such efforts have suffered from an inability to handle the large distribution shift in large accelerator facilities whose many components are time varying, and have suffered from a lack of physics constraints. In this work, we present an adaptive machine learning approach, based on convolutional neural network-based autoencoders coupled with nonlinear model-independent adaptive feedback control algorithms, and with built in hard physics constraints, towards developing robust adaptive 6D phase space diagnostics of intense charged particle beams. We show how incorporating feedback and hard physics constraints makes our approach much more robust to distribution shift as well as much more robust for extrapolation (extrapolation is a well known challenge for ML methods which typically can only handle interpolation without re-training). |
Friday, April 5, 2024 1:42PM - 1:54PM |
M17.00002: New Deep Learning based approach to Primary Vertex finding in ATLAS experiment rocky B Garg, Lauren a Tompkins In this current research, we explore the potential of deep neural networks (DNNs) for the identification and precise localization of primary vertices (PVs) within proton-proton collisions at the LHC. This innovative approach, termed as PV-Finder, adopts a hybrid methodology, initiating with kernel density estimators (KDEs), analytically calculated from a heuristic ensemble of charged track parameters, which serve as input for the DNN alongside ground truth PV information, facilitating the extraction of PV positions. The neural networks undergo rigorous training on an extensive ATLAS simulated dataset, with subsequent evaluation on an independent test dataset. To validate the efficacy of our algorithm, we conducted a comparative analysis against the standard vertex finder algorithm in ATLAS. We also share our ongoing work and future plans to develop an ‘end-to-end’ tracks-to-histograms DNN where we replace analytical calculation of KDEs with fully connected NN layers. Our research strives to make impactful contributions to the continuous evolution of data analysis and machine learning techniques within the realm of High Energy Physics. |
Friday, April 5, 2024 1:54PM - 2:06PM |
M17.00003: Quantum Machine Learning – Overview, Opportunities and Challenges J.P. Auffret The National Science Foundation in its 2023 Artificial Intelligence Research Institute Program Solicitation (NSF 23-610) highlights the potential benefit of developing AI methodologies which do not rely on “massive datasets” and the “continuing growth of data-driven models and their access to more data”. New private sector companies such as Liquid AI are exploring dynamic machine learning methodologies to decrease training data set size and computer processing requirements and increase model interpretability. |
Friday, April 5, 2024 2:06PM - 2:18PM |
M17.00004: Inferring IGM properties from EoR using neural networks Madhurima Choudhury The redshifted 21-cm line of neutral hydrogen is a sensitive probe to investigate the different phases of the evolution of our Universe. The epoch of reionization marks a crucial phase transition in the high-redshift Universe, where the neutral Hydrogen in the IGM becomes completely ionized. Observations of this 21-cm line will directly enable us to map the young Universe, over a range of cosmic times and give us deep insight into the morphology of the ionization structures which were carved out by the first sources of light, as well as about the origin and evolution of these first generation sources. Detection of the HI 21-cm power spectrum is one of the primary science drivers of several ongoing and upcoming low-frequency radio interferometers, for example: LOFAR, MWA, HERA, SKA, etc. The connection between the IGM and the measured 21-cm power spectrum is very interesting, yet it's quite non-trivial. In this work, we use Artificial Neural Networks to develop a framework which will enable us to extract and constrain IGM properties, like bubble size distribution and reionization histories from 21cm power spectrum measurements at a single redshift. |
Friday, April 5, 2024 2:18PM - 2:30PM |
M17.00005: Emulating stellar profiles for binary population synthesis with POSYDON: I can't believe it's not MESA! Elizabeth Teng, Philip Srivastava, Aggelos Katsaggelos, Zoheyr Doctor, Vicky Kalogera Understanding the internal structure of stars is crucial to modeling their evolution. The first version of the novel binary population synthesis code POSYDON includes a module for interpolating stellar and binary properties at the end of MESA evolution. In this work, we present a new profile interpolation function for predicting the internal stellar structure of density, hydrogen mass fraction, and helium mass fraction using machine learning techniques. We use principal component analysis for dimensionality reduction, and neural networks to make predictions. We find accuracy to be within the order of 10%. By delivering more information about the stars during evolution, these methods will expand the kinds of studies possible with POSYDON and other population synthesis codes, especially with respect to constraining uncertain evolutionary stages. |
Friday, April 5, 2024 2:30PM - 2:42PM |
M17.00006: EGAT: A Graph-based Tool for Reaction Property Prediction Sai Mahit Vaddadi Determining the relevance of distinct reaction pathways hinges upon the search for relevant transition states and respective barriers. Historically, such a property has been found using transition state searches based on the highest level of theory afforded. However recently, several groups have popularized the idea of directly predicting activation energies based only on reactant and product graphs using neural networks and a broad enough dataset. In this talk, we revisit this task using the Reaction Graph Depth 1 (RGD1) transition state dataset and a set of several newly developed graph attention architectures colloquially known as EGAT. Such architectures achieve state-of-the-art predictions of ~4 kcal/mol mean absolute error (MAE) on withheld test reaction sets – outperforming current architectures present. However, these architectures performed poorly on external testing sets composed of reactions with differing mechanisms, molecularity, and molecular size. Further sets of case studies on other contemporary architectures indicated that such poor out-of-sample performance is a common trait. Thus, we conclude that standard graph architectures can achieve results comparable to the irreducible error of current reaction datasets, but out-of-sample performance remains poor. |
Friday, April 5, 2024 2:42PM - 2:54PM |
M17.00007: A data-driven discovery method with limited data Himanshu Singh In mathematical framework, complex real-life models such as fluid flow across a cylinder etc. are referred as Dynamical Systems. The modeling of such complex dynamical systems is explicitly based on extracting spatio-temporal coherent structures of it. In order to comprehend these modal information, practitioners often encounters with various challenges and central to these challenges, lack of data is surely the primary one. The already established theory of Koopman and Liouville operators around reproducing kernel Hilbert spaces (RKHS) do indeed, help us in understanding them via reduced-ordered modeling algorithms, in particular, Dynamic Mode Decomposition (DMD). However, we do not have sufficient mathematical framework to understand the system when only limited data snapshots are available. |
Friday, April 5, 2024 2:54PM - 3:06PM |
M17.00008: Revolutionizing Hydrogen Generation with MXenes: The Role of Machine Learning in Designing Efficient Catalysts Abraham M Bokinala, Priyanka Sinha, Prosun Halder, Jayant K Singh Advancing scientific and technological breakthroughs in environmental remediation and energy conversion relies heavily on the continuous exploration of catalysts. Two dimensional carbides/nitrides (MXenes), with their distinctive properties are now under the spotlight for their potential in catalysis, introducing fresh possibilities. Nevertheless, the complex chemical compositions of these materials pose a challenge in identifying suitable candidates. Consequently, there is a pressing need to discover properties that provide insights into catalytic activity at a lower computational cost, particularly when navigating through related material families. The recent integration of machine learning (ML) with computational studies stands out as an effective strategy for efficiently selecting optimal catalysts tailored for diverse applications. Here, we develop a comprehensive and adaptable multistep procedure that employs various supervised ML algorithms to design well-trained data-driven models capable of predicting the activity of the hydrogen evolution reaction (HER) across a dataset comprising 4,500 MXenes. The Gradient Boosting Regressor (GBR) emerges as the most desirable ML model, demonstrating accurate and rapid predictions of the Gibbs free energy of hydrogen adsorption (ΔGH) when it is processed through recursive feature elimination (RFE), hyperparameter optimization (HO) and leave-one-out (LOO) approach. Utilizing this model, we have successfully predicted the potential of MXenes containing O functionalization with Cr, Mo and Nb metals for hydrogen generation. An analysis of feature importance reveals key descriptors governing HER performance, including the d-band centre variance with respect to the average, electron affinity and the number of valence electrons of the terminating groups. Overall, the well-trained ML model not only achieves predictive accuracy comparable to DFT calculations but also reveals the factors influencing HER activity and enables a coherent path to explore numerous configurations of MXenes. |
Friday, April 5, 2024 3:06PM - 3:18PM |
M17.00009: Coherence influx is indispensable for quantum reservoir computing Shumpei Kobayashi, Kohei Nakajima, Quoc Hoan Tran Echo state property (ESP) is a fundamental concept for an input-driven dynamical system to perform information processing tasks. Recently, the authors proposed extensions of ESP to possibly non-stationary systems and subsystems. In this paper, as a follow-up, we theoretically analyze sufficient and necessary conditions for a quantum system to satisfy non-stationary ESP and subset/subspace non-stationary ESP. Based on extensive usage of the Pauli transfer matrix (PTM) form, we find that 1) the interaction with a quantum-coherent environment, termed \textit{coherence influx}, is indispensable to realize the non-stationary ESP, and 2) the spectral radius of PTM can characterize the fading memory property of QRC. Our numerical experiment, involving a system with a Hamiltonian that entails a spin-glass/many-body localization phase, reveals that the spectral radius of PTM can describe the dynamical phase transition intrinsic to such a system. To comprehensively understand the mechanisms under ESP of QRC, we propose a simplified model: multiplicative RC (mRC), an RC system with one-dimensional multiplicative input. Theoretically and numerically, we show that the corresponding parameter to spectral radius/coherence influx in mRC directly correlates with its linear memory capacity (MC). Our findings will enlighten a theoretical aspect of PTM and the multiplicative nature of QRC. They will lead to a further understanding of QRC and information processing in open quantum systems. |
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