Bulletin of the American Physical Society
APS March Meeting 2019
Volume 64, Number 2
Monday–Friday, March 4–8, 2019; Boston, Massachusetts
Session K18: Machine Learning Material and Experimental DataFocus
|
Hide Abstracts |
Sponsoring Units: DCOMP Chair: Noam Bernstein, United States Naval Research Laboratory Room: BCEC 156B |
Wednesday, March 6, 2019 8:00AM - 8:36AM |
K18.00001: Machine learning interatomic potentials with and without (much) human labor Invited Speaker: Noam Bernstein In this talk I will show the state of the art in generating an interatomic potential using machine learning techniques (specifically, the Gaussian Approximation Potential (GAP) framework, with the Smooth Overlap of Atomic Positions (SOAP) representation of atomic geometry) that is capable of representing the Born-Oppenheimer potential energy surface of a material based on data (energies and forces) computed using density functional theory. First I will show that a careful and very human labor-intensive process of assembling a training database results in a potential with exquisite accuracy, far surpassing any empirical potential in the literature in predicting material properties. Then I will demonstrate, using several different elemental compounds with quite different bonding chemistry, that the database building process can be largely automated. A combination of iterative training and ab initio random search (AIRSS) can be used to simultaneously "discover and fit" a material without the need for any prior knowledge of what structures are relevant for a given material. |
Wednesday, March 6, 2019 8:36AM - 9:12AM |
K18.00002: Machine Learning for Ultrafast Electron Diffraction Invited Speaker: Aravind Krishnamoorthy Ultrafast electron diffraction (UED) experiments provide high-quality data about atomic structure and dynamics of functional materials down to fs-ps timescales. Imminent improvements in repetition rate of electron sources and experimental facilities will dramatically increase the size of this available data and will provide new capabilities for ultrafast science. These advances will require analysis techniques that can efficiently extract atomistic insights from raw diffraction images. In this talk, I will describe a deep generative model, trained on existing ultrafast electron diffraction data on photoexcited two-dimensional and layered materials obtained at SLAC, as well as trajectories from classical molecular dynamics and non-adiabatic quantum molecular dynamics simulations. The model is used to analyze lattice distortions, phonon modes and changes in local crystal structure due to photoexcitation and identify potential precursors to structural phase transformations in these materials. Extensions to the model to utilize streaming data for real-time analysis and the utility of the model in experimental design will also be discussed. |
Wednesday, March 6, 2019 9:12AM - 9:24AM |
K18.00003: Prediction of Molecular Properties Using Graph Kernel and Active Learning Yu-Hang Tang, Wibe A De Jong In this talk, we present a new machine learning method for training predictive models of extensive molecular properties through the application of a similarity kernel on graphical representations of molecules, which is intuitive and can adapt to molecules of arbitrary size and topology. The pairwise similarity matrices between molecules as computed by the graph kernel can be used to construct Gaussian process regression models that can predict extensive properties with provable size scaling. Using an active learning procedure, we demonstrate that models created by our method can achieve a state-of-the-art accuracy of less than 1 kcal/mol on predicting atomization energies for molecules in the QM7 dataset without using any explicit energy decomposition/localization scheme. The method also uses a much smaller number of training samples as compared to other methods to achieve the same level of accuracy. |
Wednesday, March 6, 2019 9:24AM - 9:36AM |
K18.00004: Tree Tensor Networks for Generative Modeling Song Cheng, Tao Xiang, Lei Wang, pan zhang Tensor Network States are widely used representations for many-body quantum states. They have close connections to the Graphical Models for high-dimensional data. In both research domains employing patterns such as locality or low information complexity are crucial for designing the model architecture. We employ Tree Tensor Network (TTN) for generative model. The TTN exhibits balanced performance in expressibility and efficient training and sampling. We apply TTN generative model on random binary patterns and the binary MNIST datasets and compare its performance with the matrix product states and other the popular generative models such as the Variational AutoEncoder and PixelCNN. Finally, we discuss about the future development of Tensor Network States in machine learning problems. |
Wednesday, March 6, 2019 9:36AM - 9:48AM |
K18.00005: Learning from the Density to Correct Total Energy and Forces in First Principle Simulations Sebastian Dick, Marivi Fernandez Serra We propose a new molecular simulation framework that combines the transferability, robustness and chemical flexibility of an ab initio method with the accuracy and efficiency of a machine learned force field. The key to achieve this mix is to use a standard density functional theory (DFT) simulation as a pre-processor for the atomic and molecular information, obtaining a good quality electronic density. General, symmetry preserving, atom-centered electronic descriptors are then built from this density to train a neural network to correct the baseline DFT energies and forces. These electronic descriptors encode much more information than local atomic environments, allowing a simple neural network to reach the accuracy required for the problem of study at a negligible cost. The balance between accuracy and efficiency is determined by the baseline simulation. This is shown in results where high level quantum chemical accuracy is obtained for simulations of liquid water at standard DFT cost, or where high level DFT-accuracy is achieved in simulations with a low-level baseline DFT calculation, at about one order of magnitude reduced cost. |
Wednesday, March 6, 2019 9:48AM - 10:00AM |
K18.00006: Microscopic Particle Localization Under Low-Light Conditions Shao Ran Huang, Rashid Zia Microscopic localization of particles under low-light conditions is a challenging task in microscopy, e.g. in biological physics and quantum photonics. While there has been considerable research on developing novel computational techniques, relatively less attention has been paid to integrated computational-experimental approaches leveraging hardware binning. Hardware binning allows one to judiciously sacrifice resolution, which is often excessive in locating objects of several pixels wide, to enhance the signal-to-noise ratio. Our research questions the default choice of single pixel measurements and investigates potential gains by combining multiple frames with differently hardware-binned images. Given the recent success in deep learning microscopy, we also investigate the use of deep residual learning in computing the convolution of the image with a suitable kernel for localization. |
Wednesday, March 6, 2019 10:00AM - 10:12AM |
K18.00007: Computational screening of experimental structural repositories for novel Li-ion conductors Leonid Kahle, Aris Marcolongo, Nicola Marzari A comprehensive screening of structural databases for ionic conductors by means of atomistic simulations could identify novel candidates for next-generation solid-state lithium-ion batteries, and deepen our understanding of the microscopic processes and structural motifs governing ionic diffusion in the solid state. This task is challenging because no classical simulation potential is predictive for wide variaties of materials classes, and first-principles simulations struggle to reach the necessary timescales. To model ionic diffusion efficiently and accurately, we derive a novel hybrid quantum/empirical model that can be used for molecular dynamics simulations of solid-state diffusion [1], by applying simple and intuitive approximations to fully self-consistent density-functional theory. This models underpins our high-throughput screening efforts for Li-ion conductors, powered by the AiiDA materials informatics [2] platform. I will present the different screening stages, show how high-level workflows can be used to automate and optimize the calculation of transport coefficients, and provide early results on promising candidates. |
Wednesday, March 6, 2019 10:12AM - 10:24AM |
K18.00008: Insights on materials space Chandramouli Nyshadham, Kennedy lincoln, Gus Hart Using a kernel-based machine learning surrogate model, we present insights on generating and choosing the training and testing data for optimal modeling of materials space. We introduce a tool that helps us build an “ideal” kernel, which predicts with high accuracy on small training sets. We also present a methodology for quantifying the accuracy of any kernel based surrogate model for interpolating materials space. Our insights (based on analyzing data from over 73,000 unrelaxed DFT calculations comprising 45 different materials) helped improve our model’s predictions by as much as 50% for some systems. |
Wednesday, March 6, 2019 10:24AM - 10:36AM |
K18.00009: Quantum Criticality and Possible Superconductivity in Zn Anisotropic Quantum Rotor Model Coupled with Free Fermion Jianqiao Liu, Ryuichi Shindou Quantum rotor model with Zn clock term (Zn anisotropic quantum rotor model) is known to exhibit a U(1) symmetric quantum critical point between Zn symmetry breaking phase and quantum disorder phase for n ≥ 4. Around such quantum critical point, gapped bosonic excitations in the two phases become massless. Thereby, a coupling between a local boson degree of freedom and a free fermion could lead a superconductivity near the critical point. Motivated by this anticipation, we mapped the model into an electric flux model with electric charge, and carried out a quantum Monte-Carlo simulation study with a global update. Based on a finite-size scaling analysis of a superfluid stiffness parameter and comparison with the critical exponent of the 3D XY universality class, we discuss the critical nature of the quantum critical point . If the time allows, we discuss the possible superconducting pairing instability in the fermion system. |
Wednesday, March 6, 2019 10:36AM - 10:48AM |
K18.00010: OMDB-GAP1: A new dataset for band gap predictions for large organic crystal structures Bart Olsthoorn, Richard Geilhufe, Stanislav Borysov, Alexander Balatsky Large datasets of ab initio calculations have enabled many pioneering studies of machine learning applied to quantum-chemical systems. For example, the machine learning models achieved chemical accuracy on the popular QM9 dataset which contains small organic molecules. Here, we present a new, more challenging dataset of 12,500 large organic crystal structures and their corresponding DFT band gap, freely available at https://omdb.diracmaterials.org/dataset. The dataset is based on the Organic Materials Database (OMDB) which hosts electronic properties of previously synthesized organic crystal structures. With an average of 85 atoms per unit cell, this dataset provides a new challenge for machine learning applications. We also evaluate the performance of two recent machine learning models on this new dataset: Kernel Ridge Regression with the Smooth Overlap of Atomic Positions (SOAP) and the deep learning model SchNet. |
Wednesday, March 6, 2019 10:48AM - 11:00AM |
K18.00011: Theory of band gaps in nano-porous SiC Blair Tuttle, Colton Barger, Andrew O'Hara, Sokrates T Pantelides Nano-porous SiC is an insulator used as a back-end-of-the-line dielectric in scaled integrated circuits. In the present study, nano-porous SiC atomic models are created from cubic SiC supercells. First, pores of varying sizes are created and hydrogen passivated. Then, bond switching techniques are applied to create models with variable bond densities and bond types. Density functional theory calculations are used to determine the model’s physical and electronic properties including band gaps. We apply linear regression and random forest techniques to explore the role of bonding on the bandgaps of nano-porous SiC alloys. |
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