Bulletin of the American Physical Society
APS March Meeting 2019
Volume 64, Number 2
Monday–Friday, March 4–8, 2019; Boston, Massachusetts
Session A18: Machine Learning Material and Experimental Data IFocus
|
Hide Abstracts |
Sponsoring Units: DCOMP DCMP DAMOP Chair: Jordan Venderley, Cornell University Room: BCEC 156B |
Monday, March 4, 2019 8:00AM - 8:36AM |
A18.00001: Unsupervised machine learning of single crystal x-ray diffraction data Invited Speaker: Jordan Venderley Data analysis has become a critical bottleneck in reciprocal spaces studies of single crystal x-ray diffraction. This is because while dramatic leaps in detector technology have enabled the collection of large amounts of data in short amounts of time, a parallel development in data analysis has been lacking. Current approaches for investigating this data often require highly devoted researchers to manually comb through it looking for evidence of new physics. This is time-intensive and increasingly infeasible as datasets continue to grow in size. Here we discuss an unsupervised machine learning approach to studying phase transitions in single crystal x-ray diffraction data. Our method leverages novel techniques for approximate Bayesian inference to identify contributions to the scattering intensity with distinct physical origins. It is highly scalable and employs recent developments in numerical linear algebra for memory efficiency and high speed on parallel computing hardware, such as graphical processing units. |
Monday, March 4, 2019 8:36AM - 8:48AM |
A18.00002: X-ray hyperspectral classification of the metal-insulator transition in NdNiO3 William Zheng, Alexander Swinton McLeod, Kirk W Post, Matthias Hepting, Martin Bluschke, Matteo Minola, Alexander Boris, Eva Benckiser, Rajesh V Chopdekar, Andreas Scholl, Bernhard Keimer, Dimitri Basov Rare-earth nickelates, such as NdNiO3, belong to a family of strongly correlated electron systems in which the electronic and magnetic properties are strongly coupled: NdNiO3 undergoes a temperature driven first-order Metal to Insulator phase transition (MIT) accompanied by charge-order and spin density wave-order phase transition. During the phase transition, the underlying evolution of the insulating domains are averaged out by the limited spatial resolution of conventional bulk probes but, soft X-ray photoemission electron microscopy (X-PEEM) can be applied in order to elucidate nanoscale heterogeneity. In this work we apply X-PEEM to image cooling and heating MITs and develop machine learning based analysis techniques, mainly Principle Component Analysis (PCA) and Independent Component Analysis (ICA) with k-means classification, to classify thermally evolving metallic and insulating domains. The performance of our suite of classifiers is evaluated in this novel application and we discuss their physical interpretation on the basis of charge ordering through the MIT. This class of tools can be applied to other experimental hyperspectral data to extract the characteristics of coexisting phases from otherwise intractably large datasets. |
Monday, March 4, 2019 8:48AM - 9:00AM |
A18.00003: Classifying Grazing Incidence X-ray Scattering Patterns via Convolutional Neural Networks Charles Melton, Shuai Liu, Alexander Hexemer, Daniela Ushizima Nano-structured thin films have a variety of applications, such as antireflecting coatings for solar cells, waveguides, gaseous sensors, and piezoelectric devices. Grazing-incidence small-angle X-ray scattering (GISAXS) has become a key technique to determine the morphologies of such thin films. One of the main challenges is to determine the structure information encoded in the data based on scattering patterns alone. We propose a computational scheme that learns the structure of well-defined layers of nanoparticles from GISAXS patterns. We explore this class of thin-film materials in terms of physics-based simulation models and experimental data and apply convolutional neural networks to the simulated data to obtain the encoded information of the morphology. Our classification models categorize millions of simulated scattering patterns with success rates over 94%. In addition, we show how these data-driven models have the potential to decrease analysis time of real scattering patterns from experim |
Monday, March 4, 2019 9:00AM - 9:12AM |
A18.00004: Deep learning X-ray Absorption Near Edge Spectra Liang Li, Maria Chan The interpretation of core-level spectroscopy data, such as x-ray absorption, has long been a challenging problem. In this talk, we will discuss the use of deep learning for the interpretation of x-ray absorption near edge spectra (XANES). In this work, computed spectra using a Bethe-Salpeter Equation-based approach, of transition metal oxides, are used as the training set. Corresponding experimental data of the system will be interpreted using the trained neural networks. We will discuss, in addition, the hyperparameter tuning and optimization for bias-variance tradeoff. |
Monday, March 4, 2019 9:12AM - 9:24AM |
A18.00005: Using machine learning to predict local chemical environments from X-ray absorption spectra Deyu Lu, Matthew Carbone, Mehmet Topsakal, Shinjae Yoo X-ray absorption spectroscopy is an element-specific technique for materials characterization. Specifically, X-ray absorption near edge structure (XANES) encodes important information of the local chemical environment (LCE, e.g. coordination number, symmetry and oxidation state) of the absorber atom that is key to the understanding of the electronic and chemical properties of materials. As such, unraveling the LCE from XANES spectra is akin to solving a challenging inverse problem. Existing methods rely on empirical fingerprints, which are often qualitative or semi-quantitative and not transferable. In this study, we present a machine learning-based approach to classify the LCE's of eight 3d transition metal families from the simulated K-edge XANES of a large number of compounds. The machine learning classifier can learn important spectral features in a broad energy range without human bias and once trained, can make predictions on the fly. We found that the machine learning classifier can achieve about 85% accuracy across the wide chemical space. |
Monday, March 4, 2019 9:24AM - 9:36AM |
A18.00006: Machine Learning Approach for the Discovery of Enhanced Magnetocaloric Effect in Single Molecule Magnets Ludwig Holleis, Bellave Shivaram, Prasanna Balachandran Single molecule magnets (SMM) are candidate materials for magnetocaloric applications, high-density information storage, magnetic qubits, and spintronic devices. These molecules are made of several lanthanide and/or transition metal ions coordinated by organic ligands. Despite the progress made in experimental and traditional first-principles modeling efforts, lack of predictive design guidelines hinder rapid design of SMM for targeted applications. Here, we develop a machine learning approach for predicting novel SMM for magnetocaloric applications. We construct a database from surveying the published literature on magnetocaloric effect in SMM and develop a representation scheme that include aspects related to dimensionality, structure, local coordination environment, ideal number of spins of magnetic ions, ligands, and linking chemistry. We train machine learning models to predict the entropy change. The models capture successfully the observed trends and identify key variables that contribute to the entropy. We also predict new SMMs and await experimental validation. |
Monday, March 4, 2019 9:36AM - 9:48AM |
A18.00007: A Classifier for Metal-Insulator Transitions Nicholas Wagner, James M Rondinelli We have assembled the largest dataset of resistivity-temperature measurements on temperature-activated metal-insulator transitions (MITs) to date (45 unique compounds). We supplemented this dataset with additional entries on metals and insulators with known transport behavior, i.e., do not undergo temperature-driven MITs, for comparison. We then use the 147 compounds to formulate a machine-learning model using features we collected, which describe chemical composition (e.g. mean electronegativity, atomic radii, and elemental heat of fusion); overall and local atomic structure; and estimates of the on-site electron repulsion, charge transfer energy, and compound polarizability. From this data, we constructed a machine-learning classifier to predict whether a material would undergo a MIT or not. Our model achieves a cross-validation AUC score of 88.24 +/- 11.63 and a mean accuracy of 79.23 +/- 9.23%. We also conducted a survey of 51 graduate students, faculty, and staff scientists to estimate the ability of scientists to perform this classification. The mean accuracy for humans was 59.8%. |
Monday, March 4, 2019 9:48AM - 10:00AM |
A18.00008: Machine-learning model to predict adsorption energies in thiolated bimetallic nanoclusters Gihan Panapitiya, Guillermo AvendaƱo Frano, James Patrick Lewis We developed a random forest based machine learning model to predict adsorption energies in Ag-alloyed thiolated gold nanoclusters. The features of this model are based only on the geometric properties of non-relaxed and adsorabte-free nanocluster. The features have been defined so that they can be applied to any nanocluster. Using Au25 system as a test case, we obtained prediction accuracies of 0.173 (RMSE) and 0.779 (R2). To show the applicability of our model to nanoclusters with different sizes and shapes, we also predicted adsorption energies in Au36 and Au133 nanoclusters. Our model can be used as a filtering tool to downselect nanoclusters with desired adsorption energies for further calculations. |
Monday, March 4, 2019 10:00AM - 10:12AM |
A18.00009: Model Selection Based on Bayesian Inference that Uncovers Fundamental Dynamics of Desiccation Crack Patterns Shin-ichi Ito, Akio Nakahara, Satoshi Yukawa We investigate dynamic properties of fragment size distribution in surface crack patterns observed on a thin layer of drying dense colloidal suspension experimentally and theoretically. The model selection analysis based on Bayesian inference reveals that the time-varying fragment size distribution observed in experiments exhibits a dynamic transition in its functional form from a lognormal distribution to a generalized gamma distribution. In order to explain this dynamic transition theoretically, we construct a statistical model based on an elastic theory that describes the dynamics of the shrinkage of the colloidal suspension owing to the desiccation. The statistical model predicts the existence of a characteristic length scale that determines the crossover of the dynamic transition, and reproduces the functional forms of fragment size distributions observed in experiments quantitatively. |
Monday, March 4, 2019 10:12AM - 10:24AM |
A18.00010: A machine-learning approach to magnetic neutron scattering Robert Twyman, Stuart J. Gibson, James Molony, Jorge Quintanilla One of the benefits of magnetic neutron scattering (NS) is that it can constrain the parameters of a pre-existing model Hamiltonian. Here we propose an unbiased approach that can be used before a model has been formulated. It combines Principal Component Analysis (PCA) with an artifical neural network (ANN). The PCA algorithm extracts the essential variables describing a set of NS cross-sections in an unsupervised way. The ANN then uses that information to learn to predict the cross-sections under different conditions, with supervision. To test our method, we apply it to simulated diffuse NS cross-sections obtained by exact diagonalisation of a previously-studied model of molecular magnets [H. R. Irons et al., PRB 96, 224408 (2017)]. Our main result is that the PCA can efficiently "discover" the number of fundamental parameters in the problem. The principal component scores capture key elements of the Physics including entanglement transitions. The ANN can then accurately predict NS cross-sections, confirming the validity of the PCA-based description. We conclude that PCA of NS data from real materials can be a powerful tool, specifically one capable of placing severe constraints on possible model Hamiltonians. |
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