Bulletin of the American Physical Society
APS March Meeting 2022
Volume 67, Number 3
Monday–Friday, March 14–18, 2022; Chicago
Session Q43: Data Science, Artificial Intelligence and Machine LearningFocus Recordings Available
|
Hide Abstracts |
Sponsoring Units: GDS Chair: Weishun Zhong, Massachusetts Institute of Technology Room: McCormick Place W-375B |
Wednesday, March 16, 2022 3:00PM - 3:36PM |
Q43.00001: Machine Learning for tuning, controlling, and optimizing semiconductor spin qubits Invited Speaker: Dominic T Lennon Quantum computers hold the potential to bring the world into a new quantum age. However, creating and controlling quantum bits (qubits) has turned out to be challenging. In semiconductor spin qubits, the qubit is encoded in the spin degree of freedom of a quantum dot, an electrical potential trap used to confine charge carriers. These quantum dots are controlled with gate voltages which are applied to nanosized gate-electrodes. Despite the vast progress in the quality of the materials hosting these qubits, energizing and tuning semiconductor qubit systems still requires a considerable amount of experience, time, and patience because of device-to-device variations. Here we overcome this by replacing the human operator with automated, AI-based algorithms. Our approach is architecture and material agnostic, which allows us to report results in various spin qubit systems such as Germanium/Silicon core-shell nanowires, silicon finFETs, and Gallium-Arsenide quantum dots. |
Wednesday, March 16, 2022 3:36PM - 4:12PM |
Q43.00002: Machine learning with the quantum earthmover's distance Invited Speaker: Seth Lloyd The quantum earthmover's distance is the quantum analogue of classical Wasserstein-1 distance, which generalizes the Hamming distance to general probability distributions/quantum states. This talk shows that the use of quantum earthmover's distance in machine learning can avoid the problem of barren landscapes in gradient descent methods. Applications are given to state and circuit learning. |
Wednesday, March 16, 2022 4:12PM - 4:24PM |
Q43.00003: Combining machine learning with first principles to model the Curie temperature of magnetic Heusler compounds Parul R Raghuvanshi, Krishnaraj Kundavu, Bhavana Panwar, Prasun Keshri, Rohit Pathak, Amrita Bhattacharya Accurate theoretical prediction of Curie temperature (Tc) of magnetic compositions prior to their synthesis in the laboratory is a challenging but vital task, especially for their permanent magnet application. In this work, we combine first-principles density functional theory (DFT) calculations with machine learning (ML) to predict the Tc of magnetic Heusler alloys. For this purpose, we gather the experimental Tc of 105 stable magnetic full Heusler alloys (DFT calculated net magnetic moment of 2 μB per formula unit). We employ a robust descriptor set (comprising of the elemental, compound structural, and compound magnetic ones), whereby compound descriptors are calculated from our DFT calculations. We build a regression model for the Tc using a systematic ML approach, whereby an unprecedented accuracy is attained using random forest. Furthermore, we use one of the compress sensing methods (SISSO) to perform dimensionality reduction and analyze the complex interplay of the dimensions, which curiously reveal the connection between ionization potential, radius, and melting points of atoms with the Tc. |
Wednesday, March 16, 2022 4:24PM - 4:36PM |
Q43.00004: Denoising scanning tunneling microscopy images with deep learning Frederic F Joucken, John L Davenport, Zhehao Ge, Eberth Quezada-Lopez, Takashi Taniguchi, Kenji Watanabe, Jerome Lagoute, Robert A Kaindl Machine learning has been very successfully applied to the denoising of photographic images in the last decade. For scientific imaging techniques, such as transmission electron microscopy or scanning tunneling microscopy (STM), denoising is naturally very appealing but challenges must be overcome. For supervised learning, one of the main challenges is the definition of the ground truth. We have addressed this issue by training a model on simulated STM images. The simulation is based on a tight-binding model and the noise is subsequently added to the ground-truth image considering the specificity of the noise in STM imaging experiments. We apply our model to chemically-doped graphene as well as thickness-dependent imaging of graphene and discuss its performance and limitations. |
Wednesday, March 16, 2022 4:36PM - 4:48PM |
Q43.00005: Exploring non-equilibrium systems with normalizing flows Christoph Schönle, Vittorio Peano, Florian Marquardt Normalizing flows are generative invertible neural-network models that gradually map a complicated probability distribution to a simple one, e.g. a normal multi-dimensional Gaussian. They can learn to sample from an empirically observed distribution and at the same time provide an estimate for this distribution. This allows for the use of information-theoretical concepts like the Kullback-Leibler divergence to explore phase diagrams, classify trajectories in non-equilibrium systems in an unsupervised fashion, as well as efficiently obtain effective model descriptions. We apply normalizing flows to examples of equilibrium and non-equilibrium physical systems. |
Wednesday, March 16, 2022 4:48PM - 5:00PM |
Q43.00006: Discovering dynamical symmetry breaking and resonances in nonlinear systems through AI. George P Tsironis, Georgios D Barmparis For an ensemble of nonlinear systems described through the discrete non-linear Schrödinger equation that model, for instance, molecules or photonic systems, we propose a method that finds efficiently the configuration that has prescribed transfer properties. We use physics-informed (PI) machine learning (PIML) to find the parameters for the targeted energy transfer (TET) [1] of an electron (or photon) to a state and the parameters for the self-trapping (ST) [2] transition in a nonlinear dimer. We create a model containing two variables, χD and χA, representing the nonlinear terms in the donor and acceptor system states. We then introduce a data-free PI loss function as 1.0 - Pj, for the TET and as 0.5 - Pj, for the ST transition, where Pj is the probability, the electron being in the targeted state, j. By minimizing the loss function, the method recovers known results in the TET model, and recaptures the original dynamic ST transition and its dependence on initial conditions. The model is also applied to a trimer configuration, containing a linear intermediate unit, discovering new resonant paths. The proposed PIML method is general and may be used in the chemical design of molecular complexes or engineering design of quantum or photonic systems. |
Wednesday, March 16, 2022 5:00PM - 5:12PM |
Q43.00007: Sample generation for the spin-fermion model using neural networks . Georgios Stratis, Phillip E Weinberg, Tales Imbiriba, Pau Closas, Adrian E Feiguin We present a sample generation method for condensed matter systems based on machine learning. In our research, we used feedforward neural networks in conjunction with the Metropolis-Hastings algorithm to generate samples for the spin-fermion model. We compared two different neural networks and a linear model, based on the RKKY approximation, and found the neural networks outperforming the linear model. Given enough training data all models generated samples close to the true distribution. Furthermore, we present a way to leverage the neural networks and linear model trained on smaller systems to generate samples for significantly larger systems. Even though the samples generated have higher variance compared with samples generated using exact diagonalization of the full system, our results indicate that the generated samples can appropriately determine the average energy and specific heat of the full system. Lastly, we are going to discuss how the symmetries of the model can be exploited to reduce the number of data needed to train the neural networks. |
Wednesday, March 16, 2022 5:12PM - 5:24PM |
Q43.00008: Deep Bayesian Experimental Design for Quantum Many-Body Systems Leopoldo Sarra, Florian Marquardt Bayesian experimental design is a technique that allows to efficiently select measurements to characterize a physical system, by maximizing the expected information gain. Recent developments in deep neural networks and normalizing flows allow for a more efficient approximation of the posterior and thus the extension of this technique to complex high-dimensional situations. In this talk, we show how this approach holds promise for adaptive measurement strategies to characterize present-day quantum many-body platforms. In particular, we focus on arrays of coupled cavities and qubit arrays. Both represent model systems of high relevance for modern applications, like quantum simulations and computing, and both have been realized in platforms where measurement and control can be exploited to characterize and counteract unavoidable disorder. Thus, they represent ideal targets for applications of Bayesian Experimental Design. |
Wednesday, March 16, 2022 5:24PM - 5:36PM |
Q43.00009: Band gap predition of very large number of novel Van der Waals heterostructures using active learing models Marco Fronzi, Michael Ford, Dawid Winkler, Olexandr Isayev The band gap is one of the most fundamental properties of condensed matter. However, an accurate calculation of its value, which could potentially allow experimentalists to identify suitable materials for device applications, is very computationally expensive. Here, active machine learning algorithms are used to leverage a limited number of accurate density functional theory calculations to robustly predict the band gap of a very large number of novel heterostructures. Using this approach, a database of approximately 2.2 million band gap values for various novel Van der Waals heterostructures has been produced. |
Wednesday, March 16, 2022 5:36PM - 5:48PM |
Q43.00010: Identifying Pauli spin blockade using deep learning with scarce experimental data Jonas Schuff, Dominic T Lennon, Simon Geyer, David Craig, Leon Camenzind, Federico Fedele, Florian Vigneau, Andreas V Kuhlmann, Richard J Warburton, Dominik Zumbühl, Dino Sejdinovic, G. Andrew D Briggs, Natalia Ares A method to readout spin qubits encoded in quantum dot devices relies on Pauli spin blockade (PSB) for spin-to-charge conversion. PSB leads to transport features that are hard to detect even for human experts. We present a machine learning algorithm capable of automatically identifying PSB. The scarcity of PSB data is circumvented by training the algorithm with simulated data. We demonstrate our approach on a silicon fin field-effect transistor device and report an accuracy of 96% on different test devices, giving proof that the approach is robust to device variability. |
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