Bulletin of the American Physical Society
APS March Meeting 2019
Volume 64, Number 2
Monday–Friday, March 4–8, 2019; Boston, Massachusetts
Session B18: Machine Learning Material and Experimental Data IIFocus
|
Hide Abstracts |
Sponsoring Units: DCOMP DCMP DAMOP Chair: Christof Weitenberg, University of Hamburg Room: BCEC 156B |
Monday, March 4, 2019 11:15AM - 11:51AM |
B18.00001: Identifying quantum phase transitions using artificial neural networks on experimental data Invited Speaker: Christof Weitenberg Machine learning techniques such as artificial neural networks are currently revolutionizing many technological areas and have also proven successful in quantum physics applications. Here we employ an artificial neural network and deep learning techniques to identify quantum phase transitions from single-shot experimental momentum-space density images of ultracold quantum gases and obtain results, which were not feasible with conventional methods. We map out the complete two-dimensional topological phase diagram of the Haldane model and provide an accurate characterization of the superfluid-to-Mott-insulator transition in an inhomogeneous Bose-Hubbard system. Our work points the way to unravel complex phase diagrams of general experimental systems, where the Hamiltonian and the order parameters might not be known. |
Monday, March 4, 2019 11:51AM - 12:03PM |
B18.00002: Classifying Snapshots of the Doped Hubbard Model with Machine Learning Annabelle Bohrdt, Christie S Chiu, Geoffrey Ji, Muqing Xu, Daniel Greif, Markus Greiner, Eugene Demler, Fabian Grusdt, Michael Knap Quantum gas microscopes for ultracold atoms can provide high-resolution real-space snapshots of complex many-body systems. We implement machine learning to analyze and classify such snapshots of ultracold atoms, which realize the Fermi-Hubbard model on a square lattice. At half-filling, we find that machine learning successfully identifies a crossover in the character of magnetic correlations with increasing temperature, in concurrence with the peak of the uniform spin susceptibility. We then extend the approach to assess two theoretical descriptions of doped antiferromagnets: a doped quantum spin liquid and a geometric string theory describing hidden spin order. Up to intermediate doping values, our algorithm tends to classify experimental snapshots as geometric-string-like, as compared to the doped spin liquid or to experimental images at high temperatures. Our results demonstrate the potential for machine learning in processing the wealth of data obtained through quantum gas microscopy for new physical insights. |
Monday, March 4, 2019 12:03PM - 12:15PM |
B18.00003: Detecting nematic order in STM/STS data with artificial intelligence Jeremy Goetz, Yi Zhang, Michael Lawler Detecting the subtle yet phase defining features in Scanning Tunneling Microscopy and Spectroscopy(STM/STS) data remains an important challenge in quantum materials. We meet the challenge of detecting nematic order from local density of states data with supervised machine learning and artificial neural networks for the difficult scenario without sharp features such as visible lattice Bragg peaks or Friedel oscillation signatures in the Fourier transform spectrum. We train the artificial neural networks to classify simulated data of isotropic and anisotropic two-dimensional metals in the presence of disorder. The supervised machine learning succeeds only with at least one hidden layer in the ANN architecture, suggesting the classification scheme is non-linear. We apply the finalized ANN to experimental STM data on CaFe$_2$As$_2$ and it predicts nematic symmetry breaking with 99\% confidence (probability 0.99), in agreement with previous analysis. |
Monday, March 4, 2019 12:15PM - 12:27PM |
B18.00004: Revealing Patterns in Scanning Probe Microscopy Data via Machine Learning Techniques Eric Hudson, Riju Banerjee, Lavish Pabbi, Anna Binion, Kevin Crust, William Dusch Machine Learning (ML) techniques have become prevalent in many diverse fields of research, with the goal of helping extract information from large, complex datasets. Its penetration into condensed matter physics is still however relatively shallow, even for application to results from techniques such as scanning tunneling microscopy (STM), where the image-based nature of the data would naturally seem to lend itself to now standard ML investigations. Here we present results of ML techniques applied to both topographic and spectroscopic STM data, demonstrating the power of these techniques to reveal previously hidden connections between the two and hence help improve our understanding of the relationship between structure and electronic properties at the atomic scale. |
Monday, March 4, 2019 12:27PM - 12:39PM |
B18.00005: Crystal Structure Prediction by Bayesian Optimization and Evolutionary Algorithm Tomoki Yamashita, Shinichi Kanehira, Nobuya Sato, Hiori Kino, Koji Tsuda, Takashi Miyake, Tamio Oguchi Crystal structure prediction methods such as random search (RS) and evolutionary algorithm (EA) have attracted attention. Previously we have developed a searching algorithm accelerated by Bayesian optimization (BO). BO is a selection-type algorithm which can efficiently select potential candidates by machine learning. First, we compared searching efficiency among RS, EA, and BO in the small system of Si16. In each algorithm, a hundred structures were searched. The importance of random generation is found compared with evolutionary operations even in EA. RS could be the most efficient for small systems. Furthermore, we develop a hybrid algorithm of BO and EA, and discuss the searching efficiency in large systems. |
Monday, March 4, 2019 12:39PM - 12:51PM |
B18.00006: Phonon Calculations of Phase Change Materials Using Machine-Learning Methods Youngjae Choi, Wooil Yang, Seung-Hoon Jhi Machine-learning (ML) methods for constructing the potential-energy-surface (PES) have been developed and being widely applied to the systems that may be inaccessible by conventional ab initio calculations [1, 2, 3]. Dynamical properties of the systems can also be analyzed utilizing calculated PES from the ML methods [2, 4]. One issue that should be addressed for proper application of the ML methods is the transferability of the PES, in particular for calculation of the dynamical properties in various phases. In this study, we used the ML methods to calculate the PES and phonon modes of phase change materials and carried out the transferability analysis by controlling the training sets and force errors. |
Monday, March 4, 2019 12:51PM - 1:03PM |
B18.00007: Developing computationally efficient potential models by genetic programming Alberto Hernandez, Adarsh Balasubramanian, Fenglin Yuan, Tim Mueller The length and time scales of atomistic simulations are limited by the computational cost of the methods used to predict material properties. We have developed a machine learning algorithm based on genetic programming to discover computationally efficient and parsimonious potential models. Genetic programming is an evolutionary algorithm that can search the space of functional forms, facilitating the optimization of the computational efficiency without the need of selecting an expression a priori. Our approach was validated by rediscovering the Lennard Jones potential and the Sutton Chen embedded atom model from training data generated using these models. By using training data generated from density functional theory calculations, we found simple and fast potential models for elemental systems. We present our approach, the forms of the discovered models, and assessments of their transferability, accuracy and speed. |
Monday, March 4, 2019 1:03PM - 1:15PM |
B18.00008: ICA method for identifying collective modes Yadong Wu, Hui Zhai The independent component analysis method is applied to images of ultra-cold atoms. We present this model-independent method to identify the collective modes which are mixed together in a Bese-Einstein condensate from large sets of images. Machine learning method can extract features very well and ICA method can separate these mixed modes very well. |
Monday, March 4, 2019 1:15PM - 1:27PM |
B18.00009: "Perfect crime" of machine-learning potentials: 100-fold speed-up with no detectable trace of using machine learning in the final result Konstantin Gubaev, Evgeny Podryabinkin, Gus Hart, Alexander Shapeev Machine-learning interatomic potentials have been showing significant progress in accelerating atomistic modeling while preserving near DFT accuracy. To make use of such potentials, one must prepare a training dataset of atomistic configurations evaluated with DFT. This can be automated by active learning, allowing one to develop algorithms for automatically predicting materials properties with near-DFT accuracy with speedups of a few orders of magnitude. The only downside of such algorithms is a numerical error in the final answer arising from the deviation of a machine-learning potential from DFT. In my talk, I will show that, in some applications, one could develop algorithms that are free even from that numerical error. For the application of obtaining thermodynamically stable ternary alloy structures I will present an algorithm for screening out high-energy structures, thus accelerating a baseline DFT-based high-throughput algorithm by a factor of 100, leaving zero error in the final answer when compared to DFT. This alludes to the "perfect crime": machine-learning potentials offer very large speed-ups, but the final result is indistinguishable from the one obtained by pure DFT. |
Monday, March 4, 2019 1:27PM - 1:39PM |
B18.00010: Deep Learning of Lennard-Jones Potential Parameterization Alireza Moradzadeh, N. R. Aluru In this study, a deep neural network is developed to parameterize van der Waals interactions at various thermodynamic states based on the pair-correlation functions obtained through MD simulation of about 52 μs. After training, the network not only performs with high accuracy and fidelity for van der Waals particles but also performs well for correlations obtained from all-atom MD simulation of complex molecules. The network is capable of developing coarse-grained force fields within the theoretical limitation and accuracy imposed by van der Waals interactions. The accuracy and fidelity of the method are investigated by computing the total variation in the radial distribution function and the Kullback-Leibler divergence for the coarse-grained model development, while the mean-squared error is used to characterize the performance for the vdW particles. Our results show that deep learning is able to obtain the solution to inverse-problem of liquid-state theory under the assumption of a predetermined pair potential in both all-atom and coarse-grained models with a computational cost that is several orders of magnitude faster than other available methods in the literature. |
Monday, March 4, 2019 1:39PM - 1:51PM |
B18.00011: A direct and local deep learning model for atomic forces in solids Natalia Kuritz, Goren Gordon, Amir Natan We demonstrate a direct and local Deep Learning (DL) model for atomic forces. We apply this model for bulk aluminum, silicon and sodium and show that the model errors are comparable to other state of the art algorithms. Our model allows the calculation of forces in large cells using a training data that we built from smaller cells that were calculated with Density Functional Theory (DFT). In addition, we examine the question of temperature transferability of the model and show that we can train the model with data that was produced at a high temperature and then test it on data that was produced at lower temperatures. We also explore the physical properties of the system (e.g. number of nearest neighbors) effect on the model convergence with respect to some of its parameters. Finally, we discuss why the performance of such local models is better in some materials in comparison to others. |
Monday, March 4, 2019 1:51PM - 2:03PM |
B18.00012: Machine Learning Correlates CDW Properties with Local Gap in Cuprates Kaylie Hausknecht, Tatiana Webb, Michael C Boyer, Yi Yin, Takeshi Kondo, Tsunehiro Takeuchi, Hiroshi Ikuta, Eric Hudson, Jennifer Hoffman With the advent of atomic resolution imaging techniques comes the challenge of disentangling the intrinsic electronic properties of materials from their stochastic atomic-scale disorder. In the past decade, machine learning (ML) image analysis techniques have rapidly evolved, and their applications in physics are just emerging. Here, we use ML to test correlation hypotheses between spatially resolved measurements of disordered materials to overcome the limitations of standard Fourier analysis techniques. We apply artificial neural networks to uncover the doping-dependence of the density wave (DW) structure in the cuprate superconductor (Pb,Bi)2(Sr,La)2CuO6+δ (Bi-2201) imaged via scanning tunneling microscopy. In Bi-based cuprates, the electronic inhomogeneity, caused by local variations in doping, limits the precision of DW wavevector measurements. Our ML algorithm overcomes this limitation and allows clear differentiation between commensurate and incommensurate DW instabilities with physically distinct mechanisms. More broadly, our work lays the foundation for a ML approach to quantify intrinsic periodic order and correlations from datasets where these trends are masked by disorder. |
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