Bulletin of the American Physical Society
APS March Meeting 2022
Volume 67, Number 3
Monday–Friday, March 14–18, 2022; Chicago
Session S47: Machine Learning for Quantum Matter IIFocus Recordings Available
|
Hide Abstracts |
Sponsoring Units: DCOMP GDS DMP Chair: Arunkumar Rajan, Georgia Institute of Technology Room: McCormick Place W-470B |
Thursday, March 17, 2022 8:00AM - 8:36AM |
S47.00001: Towards interpretable and reliable machines learning physics Invited Speaker: Anna Dawid Machine learning (ML) models are known for their black-box construction, i.e., they usually hinder any insight into the reasoning behind their predictions. This opaqueness increases with the model complexity and expressivity and is inherent to neural networks. As a result, neither we can fully trust their predictions (lack of reliability) nor learn what the ML model learned (lack of interpretability). Both qualities are crucial, especially if we want to apply ML to novel problems with unknown solutions. Instead of limiting ourselves to simple interpretable models, we can look at the neural network through the lens of its Hessian which encodes valuable information on the model. We show how the Hessian-based methods can extract a concept of similarity between input data learned by a network, estimate the uncertainty of model predictions, or judge whether a network makes an extrapolating guess instead of a data-based decision. We present these gains on the example of convolutional neural networks learning quantum phases both in the numerically simulated and experimental data. |
Thursday, March 17, 2022 8:36AM - 8:48AM |
S47.00002: Direct sampling of projected entangled-pair states Tom Vieijra, Jutho Haegeman, Frank Verstraete, Laurens Vanderstraeten Direct sampling of high-dimensional probability distributions provides an important improvement over Markov Chains in Variational Monte Carlo algorithms for many-body quantum systems. They alleviate the burden of large autocorrelation times in Markov Chains by providing independent samples drawn according to the probability distribution. On one hand, probability distributions can be designed such that they can be easily sampled directly, e.g. in the case of autoregressive neural networks. On the other hand, a given distribution can in some cases be cast in a form such that they can be sampled directly. We show how to do this approximately for projected entangled-pair states (PEPS) by adding an importance sampling step. We show that this procedure is efficient and provides signifficant improvements over the widely used Markov Chain sampling of PEPS with local updates. |
Thursday, March 17, 2022 8:48AM - 9:00AM |
S47.00003: Neural Network Ansatz for Finite Temperature Filippo Vicentini, Riccardo Rossi, Giuseppe Carleo Neural-network Quantum States [1] are an efficient ansatz for approximating the ground and excited-states of highly entangled systems in 1 and 2 dimensional quantum systems at zero temperature. Some extensions to systems coupled to thermal baths have been recently proposed, but they are either limited to classical systems, to shallow networks encoding a Neural Density Matrix or they require 3 different sampling procedures [2]. |
Thursday, March 17, 2022 9:00AM - 9:12AM |
S47.00004: Neural network representation for minimally entangled typical thermal state Hongwei Chen, Douglas G Hendry, Adrian E Feiguin We investigate a new approach for modeling finite temperature quantum systems by generating and sampling minimally entangled typical thermal state(METTS). A restricted Boltzmann machine (RBM), a type of artificial neural network, is used as a variational wave function to represent METTS in variational Monte Carlo. We evolve the parameters to match the imaginary time evolution and calculated the thermal average of physical observables. Our results demonstrate that the properties of finite temperature quantum systems can be precisely explored by variational Monte Carlo methods. |
Thursday, March 17, 2022 9:12AM - 9:24AM |
S47.00005: Ground-state properties via machine learning quantum constraints Pei-Lin Zheng, Si-Jing Du, Yi Zhang Ground-state properties are central to our understanding of quantum many-body systems. At first glance, it seems natural and essential to obtain the ground state before analyzing its properties; however, its exponentially large Hilbert space has made such studies costly, if not prohibitive, on sufficiently large system sizes. Here, we propose an alternative strategy based upon the expectation values of an ensemble of selected operators and the elusive yet vital quantum constraints between them, where the search for ground-state properties simply equates to classical constrained minimization. These quantum constraints are generally obtainable via sampling then machine learning on a large number of systematically consistent quantum many-body states. We showcase our perspective on 1D fermion chains and spin chains for applicability, effectiveness, and several unique advantages, especially for strongly correlated systems, thermodynamic-limit systems, property designs, etc. |
Thursday, March 17, 2022 9:24AM - 9:36AM |
S47.00006: Gauge invariant autoregressive neural network for quantum lattice models Zhuo Chen, Di Luo, Kaiwen Hu, Zhizhen Zhao, Vera M Hur, Bryan K Clark Gauge symmetries arise in various aspects of quantum mechanics, from condensed matter physics to high energy physics. We develop autoregressive neural networks that explicitly incorporate gauge symmetries and algebraic constraints and allow for efficient sampling. We analytically construct the gauge invariant neural network representation of the ground and excited states of the 2D and 3D toric codes, and the X-cube fracton model. We variationally optimize our neural networks to simulate the dynamics of the quantum link model of U(1) lattice gauge theory, determine the phase transition for the 2D Z2 gauge theory, obtain the phase diagram and compute the central charge of the SU(2)3 anyonic chain. Our approach provides a framework to construct neural networks with symmetries, and shows a powerful method for exploring condensed matter physics, high energy physics and quantum information science. |
Thursday, March 17, 2022 9:36AM - 9:48AM |
S47.00007: Looking Under the Hood: How Convolutional Neural Networks Successfully Approximate Quantum Spin Hamiltonians Shah Saad Alam, Yilong Ju, Jonathan Minoff, Fabio Anselmi, Ankit B Patel, Han Pu Convolutional neural networks (CNNs) have been employed alongside Variational Monte Carlo methods for finding the ground state of quantum spin Hamiltonians. In order to do so, however, a CNN with linearly many variational parameters has to successfully approximate a wavefunction on an exponentially large Hilbert space. In our work, we look under the black box to understand how the CNN optimizes learning for spin systems, and the role played by physical symmetries during training. We then also demonstrate a method for using the symmetries of the underlying spin system to propose an improved training algorithm. Finally, to further investigate how the CNN replicates the essential physics of the target Hamiltonian, we show connections between a one-layer CNN wavefunction ansatz to ansatz from maximum entropy (MaxEnt) distribution as well as to entangled plaquette ansatzes, thus connecting the neural network to concepts from information theory and previous physics VQMC methods. |
Thursday, March 17, 2022 9:48AM - 10:00AM |
S47.00008: A first principles informed machine learning model for helical nanostructures Amartya S Banerjee, Susanta Ghosh, Shashank Pathrudkar, Hsuan Ming Yu Helical nanostructures constitute an important class of low-dimensional matter and include nanotubes of arbitrary chirality, nanowires, nanoribbons, nanoassemblies and other miscellaneous chiral structures. The fascinating electronic, optical and magnetic properties of these materials offer unparalleled opportunities for impacting the design of novel quantum, photonic and electromagnetic devices. We present a first-principles informed machine learning model that can predict the electronic structure of such materials in their natural or distorted states, while they are being subjected to deformation modes such as torsion and extension/compression. The model includes structural symmetries, atomic relaxation effects and uses a symmetry-adapted version of Kohn-Sham Density Functional Theory in helical coordinates to generate the input data. We use armchair single wall carbon nanotubes as a prototypical example, and demonstrate the use of the model to predict various electronic fields when the radius of the nanotube, its axial stretch, and the twist per unit length are specified as inputs. Our model is likely to find applications in areas where the interplay of strain and electronic properties at the nanoscale (i.e., strain engineering) plays an important role. |
Thursday, March 17, 2022 10:00AM - 10:12AM |
S47.00009: Machine learning frequency-resolved phonon transport from ultrafast electron diffraction Zhantao Chen, Nina Andrejevic, Tongtong Liu, Xiaozhe Shen, Thanh Nguyen, Nathan C Drucker, Mingda Li Developing reliable measurement techniques for frequency-resolved phonon transport is one of the central problems in thermal science and engineering. Significant efforts have been devoted to various laser-based pump-probe techniques; however, their capabilities are limited by several bottlenecks. Here, we present a machine learning-enabled computational framework that can reveal microscopic phonon transport in heterostructures and extract frequency-dependent transport properties. Using the phonon Boltzmann transport equation (BTE) in conjunction with the adjoint-state method and automatic differentiation, we learn the phonon properties that generate a particular ultrafast electron diffraction (UED) signal. This allows us to recover frequency-dependent interface transmission coefficients, the critical incident angle at the interface, and layer-specific emissivity, from which real-time and real-space phonon dynamics can be reconstructed. We demonstrate the validity of our approach on various synthetic data and apply it to experimental measurements of an Au/Si heterostructure. Our work provides a novel approach to explore phonon transport mechanisms at the nanoscale. |
Thursday, March 17, 2022 10:12AM - 10:24AM |
S47.00010: Interpretable Machine Learning for Materials Design Timur Bazhirov, James Dean, Rahul Bhowmik, Sergey Barabash, Matthias Scheffler, Thomas A Purcell Fueled by the widespread adoption of Machine Learning and the high-throughput screening of materials, the data-driven approach to materials design has asserted itself as a robust and powerful tool for the in-silico prediction of materials properties. Researchers often face a difficult choice between a model’s interpretability or its performance. We study this trade-off by leveraging four different state-of-the-art Machine Learning techniques: XGBoost, SISSO, Roost, and TPOT for the prediction of structural and electronic properties of perovskites and 2D materials. We identify key problems to address to continue down a path towards automation. Finally, we offer several possible solutions to these challenges with a focus on retaining interpretability and share our thoughts on magnifying the impact of Machine Learning on materials design. |
Thursday, March 17, 2022 10:24AM - 10:36AM |
S47.00011: Improving the Accuracy and Efficiency of Nonlocal Exchange Functionals via Machine Learning Kyle Bystrom, Boris Kozinsky Machine Learning (ML) is a promising approach to improve the accuracy of exchange-correlation (XC) functionals for Density Functional Theory (DFT). In this talk, we summarize recent developments in the CIDER formalism, an approach for developing ML exchange functionals that leverages nonlocal information about the density while allowing physical constraints on the functional to be enforced. In particular, we discuss how the features used in the model have been adjusted to improve their transferability and minimize spurious dependence of the valence electrons on the core electron distribution. We also investigate how the accuracy and transferability of the Gaussian Process ML models can be optimized by training to both the exchange energy densities and the total exchange energies of molecules. Lastly, we explore the efficient implementation of CIDER models in plane-wave DFT codes through an auxiliary basis approach similar to that used for van der Waals functionals. |
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