Bulletin of the American Physical Society
APS March Meeting 2023
Volume 68, Number 3
Las Vegas, Nevada (March 5-10)
Virtual (March 20-22); Time Zone: Pacific Time
Session D53: Machine Learning for SpectroscopyFocus
|
Hide Abstracts |
Sponsoring Units: GDS Chair: Nina Andrejevic, Argonne National Laboratory; Davis Unruh, Argonne National Laboratory Room: Room 307 |
Monday, March 6, 2023 3:00PM - 3:36PM |
D53.00001: Teaching Core-Hole Spectroscopy to a Deep Neural Network Invited Speaker: Conor Rankine Deep neural networks (DNNs) – multilayer machine-learning models that are able to extract and learn patterns represented in data without hand-coded heuristics – are transforming what we can do, and the way we do it, across the physical sciences. |
Monday, March 6, 2023 3:36PM - 3:48PM |
D53.00002: AutoML-accelerated EELS/XAS as an advanced structure characterization tool Haili Jia, Gihyeok Lee, Yiming Chen, Wanli Yang, Maria K Chan Atomistic structures of materials provide great insights of the functionality of the materials. Determining atomic structures is a fundamental problem in materials science. Although there are both experimental and computational methods to determine these nanoscale structures, they both possess limitations which demonstrate the difficulty of nanoscale structural determination, especially for systems with defects. We aim to tackle this problem and conduct structure characterization by combining ab initio simulations, experimental acquisition, and machine learning (ML) techniques. While ML methods have been widely applied in materials characterization, these frameworks crucially rely on human machine learning experts to perform manual tasks. In this work, we seek to predict the local structures and properties from real spectroscopic data (EELS/XAS) in an automated fashion based on multi-fidelity Bayesian optimization, which includes the conversion between simulation data to and experimental data, feature engineering, hyperparameter tuning and ML model selection. For the material system, we use a variety of lithium nickel manganese cobalt oxides (NMC) compounds as the test cases, including those with oxygen vacancies and antisite defects. We will demonstrate that our framework can not only accurately predict the target information, but also provide the interpretability which quantitatively bridges the spectroscopy with the local atomic and electronic structures around the adsorbing atom. |
Monday, March 6, 2023 3:48PM - 4:00PM |
D53.00003: Featurization Approaches for Machine Learning of X-ray Absorption Spectra Yiming Chen, Maria K Chan, Shyue Ping Ong, Chengjun Sun, Steve M Heald, chi chen Machine learning (ML) has been accelerating the analysis and interpretation of materials characterization data. For example, it has been utilized to extract essential material properties such as oxidation states and structural information from X-ray absorption spectroscopy (XAS). While most ML models focus on the raw spectra intensities as the model input, transformation of spectra that can potentially enhance model performance remains scarcely explored. In this presentation, we will benchmark both the reduced dimension features and overcomplete representation to unveil the optimal representation of spectroscopy data. The system of interest is a cathode material for Li-ion battery, LiNixMnyCozO2 (NMC). Despite its high energy density, excellent long-term cyclability and relatively low economic cost, the transition metal mixing makes it challenging to investigate the detailed changes during electrochemical cycling. The performance of these input transformations will be assessed for XAS based on both regression and classification tasks. We will demonstrate that such featurization can significantly improve not only the prediction accuracy, but also the interpretability of ML models. Model validation on unseen experimental dataset will also be discussed to prove the model transferability. On the other hand, despite that defect of NMC materials is well-observed during experiment, less emphasis has placed on its formation mechanism and impact on battery performance. In this project, we will also aim to tackle the formation mechanism of those defects through a combination of first-principles calculations and ML techniques. |
Monday, March 6, 2023 4:00PM - 4:12PM |
D53.00004: Multi-code Benchmark on Ti K-edge X-ray Absorption Spectra of Ti-O Compounds Fanchen Meng, Benedikt Maurer, Fabian Peschel, Sencer Selcuk, Mark S Hybertsen, Xiaohui Qu, Christian W Vorwerk, Claudia Draxl, John Vinson, Deyu Lu X-ray absorption spectroscopy (XAS) is an element-specific materials characterization technique that is sensitive to structural and electronic properties. First-principles simulated XAS has been widely used as a powerful tool to interpret the experimental spectra and draw physical insights. Recently, there has also been growing interest in building computational XAS databases to enable data analytics and machine learning applications. However, there are non-trivial differences among several commonly used XAS simulation codes, both in underlying formalism and technical implementation. Developing reliable and reproducible computational XAS databases calls for systematic benchmark studies. |
Monday, March 6, 2023 4:12PM - 4:24PM |
D53.00005: Deep Learning and Infrared Spectroscopy: Representation Learning with a β-Variational Autoencoder Michael Grossutti, John R Dutcher Infrared (IR) spectra contain detailed and extensive information about the chemical composition and bonding environment in a sample. However, this information is difficult to extract from complex heterogeneous systems because of overlapping absorptions due to different generative factors. We implement a deep learning approach to study the complex spectroscopic changes that occur in cross-linked polyethylene (PEX-a) pipe by training a β-variational autoencoder (β-VAE) on a database of PEX-a pipe spectra. We show that the β-VAE outperforms principal component analysis (PCA) and learns interpretable and independent representations of the generative factors of variance in the spectra. We apply the β-VAE encoder to a hyperspectrum of a crack in the wall of a pipe to evaluate the spatial distribution of these learned representations. This study shows how deep learning architectures like β-VAE can enhance the analysis of spectroscopic data of complex heterogeneous systems. |
Monday, March 6, 2023 4:24PM - 4:36PM |
D53.00006: A machine learning framework for Raman spectrum prediction Nina Andrejevic, Michael J Davis, Mingda Li, Maria K Chan Raman spectroscopy captures materials’ vibrational properties in the form of highly resolved fingerprints with characteristic peaks. However, connecting Raman spectra to underlying structural and chemical attributes can be nontrivial and computationally expensive. In this work, we apply machine learning methods to obtain Raman spectra from accessible structural and atomic properties. Using an objective function inspired by optimal transport, we first learn low-dimensional representations of Raman spectra which serve as effective prediction targets for machine learning from materials attributes. Due to limited available training data, we employ symmetry-constrained Euclidean neural networks, which have demonstrated success on related property prediction tasks, and evaluate our framework on both ab initio and experimental spectra with varying complexity. Our approach enables rapid prediction of Raman spectra from structures which accelerates the interpretation of Raman spectroscopy data. |
Monday, March 6, 2023 4:36PM - 4:48PM |
D53.00007: AI-powered biotechnology platform of single-cell Raman micro-spectroscopy enables high-resolution dynamical phenotyping study of bacterial growth and cellular heterogeneity Zijian Wang, Jenny Kao-Kniffin, Eric J Craft, Matthew C Reid, Andrea Giometto, Kilian Q Weinberger, April Z Gu Progresses in single-cell Raman micro-Spectroscopy (SCRS) technologies and automated measurements rapidly yield large spectroscopic datasets that require advanced artificial intelligence (AI)-powered data analytics to discover biological mechanisms in high-resolution and non-invasive manner. We developed this AI-powered SCRS biotechnology platform RamanomeSpec with 4 modules to uncover bacterial phenotypic dynamics, cellular heterogeneity, and identification using machine and deep learning algorithms. |
Monday, March 6, 2023 4:48PM - 5:00PM |
D53.00008: EllipsoNet: Deep-learning-enabled optical ellipsometry for complex thin films Ziyang Wang Optical spectroscopy is indispensable for research and development in nanoscience and nanotechnology, microelectronics, energy, and advanced manufacturing. Advanced optical spectroscopy tools often require both specifically designed high-end instrumentation and intricate data analysis techniques. Beyond the common analytical tools, deep learning methods are well suited for interpreting high-dimensional and complicated spectroscopy data. They offer great opportunities to extract subtle and deep information about optical properties of materials with simpler optical setups, which would otherwise require sophisticated instrumentation. In this work, we propose a computational ellipsometry approach based on a conventional tabletop optical microscope and a deep learning model called EllipsoNet. Without any prior knowledge about the multilayer substrates, EllipsoNet can predict the complex refractive indices of thin films on top of these nontrivial substrates from experimentally measured optical reflectance spectra with high accuracies. This task was not feasible previously with traditional reflectometry or ellipsometry methods. Fundamental physical principles, such as the Kramers-Kronig relations, are spontaneously learned by the model without any further training. This approach enables in-operando optical characterization of functional materials within complex photonic structures or optoelectronic devices. |
Monday, March 6, 2023 5:00PM - 5:12PM |
D53.00009: Exploiting Sparsity in Artificial Neural Networks for Spectroscopic Data Jakub Vrabel, Erik Kepes, Pavel Nedelnik, Pavel Porizka, Jozef Kaiser Models based on Artificial Neural Networks (ANNs) are widely used in spectroscopy for various tasks (classification, regression, dimension reduction), often with state-of-the-art performances. Commonly utilized ANNs have hundreds of thousands to millions of weights, resulting in a huge over-parametrization. While the over-parametrization seems to be beneficial for the performance (e.g., double descent behavior) and generalizability, it significantly worsens the interpretability of the model. Such a black-box behavior of the models limits the applicability in high-stake applications and slows scientific progress. In this work, we exploit lottery tickets (i.e., iteratively pruned, sparse networks with the same or slightly better performance than their dense counterparts) for the interpretability of ANNs that were trained for classification and regression on spectroscopic data. We show that lottery tickets in a contrastive regime (where we compare two sparse models) can detect task-important features in the data and allow for better model interpretability. The concept is demonstrated on Laser-Induced Breakdown Spectroscopy data but can be extended to other techniques, considering the availability of a sufficient amount of data and similar properties. The results are critically evaluated and compared to a baseline approach, the feature visualization by an input optimization technique. |
Monday, March 6, 2023 5:12PM - 5:24PM |
D53.00010: Deep machine learning the spectral function of a hole in a quantum antiferromagnet Weiguo Yin, Jackson Lee, Matthew R Carbone Understanding charge motion in a background of interacting quantum spins is a basic problem in quantum many-body physics. The most extensively studied model for this problem is the so-called t-t'-t''-J model, where the determination of the parameter t' in the context of cuprate superconductors was inconclusive. Here we present a theoretical study of the spectral functions of a mobile hole in the t-t'-t''-J model using a classical machine learning (ML) method, namely K-nearest neighbors (KNN), and a deep ML method, namely fully connected feed-forward neural network (FFNN). We employ the self-consistent Born approximation to generate the training, validation, and testing dataset consisting of about 1.3x10^5 spectral functions and introduce an algorithm that reduces the ML dimensionality by 25%. We show that for the forward problem, both ML methods allow for accurate spectral functions to be calculated in significantly less time than the physical theory that produces the data, allowing for rapid search through parameter space. Furthermore, we find that for the inverse problem, FFNN can, but KNN cannot, accurately predict the model parameters using merely the density-of-state spectrum. Our results suggest that it may be possible to use deep learning methods to predict material parameters from experimentally measured spectral functions. |
Monday, March 6, 2023 5:24PM - 5:36PM Author not Attending |
D53.00011: A Modernized View of Coherence Pathways in Magnetic Resonance Spectroscopy John M Franck Most Magnetic Resonance Experiments involve the acquisition of very large datasets consisting of complex data. Since relatively complicated Nuclear Magnetic Resonance (NMR) experiments date back decades, the traditional approach to this data was to dramatically reduce the amount of data at the accumulation stage. In particular, cycling the phase of excitation pulses allows one to isolate signal that arises from particular coherence pathways (i.e. signal that pulses move between particular elements of the density matrix following a particular pattern/pathway), and methods were developed to accumulate the data in a way that isolates the data from a particular pathway while discarding data from undesired pathways. |
Monday, March 6, 2023 5:36PM - 5:48PM |
D53.00012: Machine Learning for Improvements to Gamma Spectroscopy in Nuclear Fusion Diagnostics Kimberley S Lennon, Callum Grove, Joseph Neilson, Chantal Nobs, Lee Packer, Robin Smith Fusion diagnostics are critical on the path to commercial fusion reactors, since the ability to understand and measure plasma features is important to sustaining fusion reactions. Gamma spectroscopy is one technique used to aid fusion diagnostics, to provide information on ion distribution and also in neutron activation analysis to calculate fusion power. However, a common feature with gamma spectroscopy is Compton scattering events within the detector. These elevate the background, reducing the likelihood of detecting peaks from low-energy gamma rays, leading to higher detection and characterisation limitations. |
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