A5041716995- Advanced Condensed Matter Physics
- Quantum, superfluid, helium dynamics
- Magnetic and transport properties of perovskites and related materials
- High-pressure geophysics and materials
- Physics of Superconductivity and Magnetism
- Nuclear Physics and Applications
- Rare-earth and actinide compounds
- Material Dynamics and Properties
- Hydrogen Storage and Materials
- Advanced Chemical Physics Studies
- Nuclear Materials and Properties
- Optical and Acousto-Optic Technologies
- Spectroscopy and Quantum Chemical Studies
- 2D Materials and Applications
- Domain Adaptation and Few-Shot Learning
- Solid-state spectroscopy and crystallography
- Carbon Nanotubes in Composites
- Multiferroics and related materials
- Multimodal Machine Learning Applications
- Iron-based superconductors research
- Glass properties and applications
- Advanced Neural Network Applications
- Nanopore and Nanochannel Transport Studies
- Topological Materials and Phenomena
- Nuclear reactor physics and engineering
Oak Ridge National Laboratory
2015-2024
Wake Forest University
2024
M.N. Mikheev Institute of Metal Physics
2024
Ural Federal University
2024
Perm State University
2024
Google (United States)
2019-2024
Joint Institute for Nuclear Research
2024
University of Salzburg
2024
Brain (Germany)
2019-2023
Google (Switzerland)
2023
A combination of density functional theory (DFT) calculations and experiments is used to shed light on the relation between surface structure Li-ion storage capacities following functionalized two-dimensional (2D) transition-metal carbides or MXenes: Sc2C, Ti2C, Ti3C2, V2C, Cr2C, Nb2C. The are found strongly depend nature groups, with O groups exhibiting highest theoretical capacities. MXene surfaces can be initially covered OH removable by high-temperature treatment reactions in first...
This work tackles the problem of semi-supervised learning image classifiers. Our main insight is that field can benefit from quickly advancing self-supervised visual representation learning. Unifying these two approaches, we propose framework (S4L) and use it to derive novel classification methods. We demonstrate effectiveness methods in comparison both carefully tuned baselines, existing then show S4L be jointly trained, yielding a new state-of-the-art result on ILSVRC-2012 with 10% labels.
Quasi-one-dimensional water encapsulated inside single-walled carbon nanotubes, here referred to as nanotube water, was studied by neutron scattering. The results reveal an anomalously soft dynamics characterized pliable hydrogen bonds, anharmonic intermolecular potentials, and large-amplitude motions in water. Molecular simulations consistently describe the observed phenomena propose structure of which comprises a square-ice sheet wrapped into cylinder interior molecules chainlike configuration.
In two dimensional honeycomb ferromagnets, bosonic magnon quasiparticles (spin waves) may either behave as massless Dirac fermions or form topologically protected edge states. The key ingredient defining their nature is the next-nearest neighbor Dzyaloshinskii-Moriya (DM) interaction that breaks inversion symmetry of lattice and discriminates chirality associated spin-wave excitations. Using inelastic neutron scattering, we find spin waves insulating ferromagnet CrI$_3$ ($T_C=61$ K) have...
Herein we show that hydrazine intercalation into 2D titanium carbide (Ti3C2-based MXene) results in changes its surface chemistry by decreasing the amounts of fluorine, OH groups and intercalated water. It also creates a pillaring effect between Ti3C2Tx layers pre-opening structure improving accessability to active sites. The treated material has demonstrated greatly improved capacitance 250 F g(-1) acidic electrolytes with an excellent cycling ability for electrodes as thick 75 μm.
Vision Transformers (ViT) have been shown to attain highly competitive performance for a wide range of vision applications, such as image classification, object detection and semantic segmentation. In comparison convolutional neural networks, the Transformer's weaker inductive bias is generally found cause an increased reliance on model regularization or data augmentation ("AugReg" short) when training smaller datasets. We conduct systematic empirical study in order better understand...
Effective scaling and a flexible task interface enable large language models to excel at many tasks. We present PaLI (Pathways Language Image model), model that extends this approach the joint modeling of vision. generates text based on visual textual inputs, with performs vision, language, multimodal tasks, in languages. To train PaLI, we make use pre-trained encoder-decoder Vision Transformers (ViTs). This allows us capitalize their existing capabilities leverage substantial cost training...
We propose a simple pairwise sigmoid loss for imagetext pre-training. Unlike standard contrastive learning with softmax normalization, the operates solely on image-text pairs and does not require global view of similarities normalization. The simultaneously allows further scaling up batch size, while also performing better at smaller sizes. With only four TPUv4 chips, we can train Base CLIP model 4k size Large LiT 20k latter achieves 84.5% ImageNet zero-shot accuracy in two days. This...
The scaling of Transformers has driven breakthrough capabilities for language models. At present, the largest large models (LLMs) contain upwards 100B parameters. Vision (ViT) have introduced same architecture to image and video modelling, but these not yet been successfully scaled nearly degree; dense ViT contains 4B parameters (Chen et al., 2022). We present a recipe highly efficient stable training 22B-parameter (ViT-22B) perform wide variety experiments on resulting model. When evaluated...
Recently, Co-based honeycomb magnets have been proposed as promising candidate materials to host the Kitaev spin liquid state. One of front-runners is BaCo$_2$(AsO$_4$)$_2$ (BCAO), where it was suggested that exchange processes between Co$^{2+}$ ions via surrounding edge-sharing oxygen octahedra could give rise bond-dependent interactions. In this work, we present and analyze comprehensive inelastic neutron scattering studies BCAO with fields in plane. Combining constraints from magnon...
A fine resolution chopper spectrometer (SEQUOIA) recently received first neutrons at the SNS. The commissioning phase of instrument is underway. SEQUOIA designed to utilize an incident energy (Ei) between 10-2000 meV. monochromatic beam provided on a sample, 20 m from decoupled ambient temperature H2O moderator, by filtering white with Fermi located 18 source. After interacting are detected array 3He linear position sensitive tubes vertical cylinder radius 5.5 m. This contribution presents...
We report inelastic neutron scattering measurements of the phonon density states in Mg 11B2, which has a superconducting transition at 39.2 K. The acoustic phonons extend energy to 36 meV, and there are highly dispersive optic branches peaking 54, 78, 89, 97 meV. A simple Born-von Kàrmàn model reproduces mode energies, provides an estimate electron-phonon coupling lambda approximately 0.9. Furthermore, estimated boron magnesium contributions isotope effect qualitative agreement with...
Using neutron scattering and ab initio simulations, we document the discovery of a new "quantum tunneling state" water molecule confined in 5 Å channels mineral beryl, characterized by extended proton electron delocalization. We observed number peaks inelastic spectra that were uniquely assigned to quantum tunneling. In addition, momentum distribution was measured with deep scattering, which directly revealed coherent delocalization protons ground state.Received 18 November...
The Spallation Neutron Source at Oak Ridge National Laboratory now hosts four direct geometry time-of-flight chopper spectrometers. These instruments cover a range of wave-vector and energy transfer space with varying degrees neutron flux resolution. regions reciprocal available to measure these are not exclusive overlap significantly. We present comparison the capabilities this instrumentation, conducted by data mining instrument usage histories, specific scanning regimes. In addition, one...
Studying the vibration of atoms is fundamental importance and can provide critical insight for understanding materials behavior, such as structure phase transition, thermodynamics, chemical reactions. The atomic be probed using vibrational spectroscopy with various incident particles photons, neutrons, or electrons. A major challenge when applying these techniques often how to interpret spectra make connections theory. To this end, methods that simulate from atomistic models are highly...
MXenes are a new class of two-dimensional materials with properties that make them important for applications include batteries, capacitive energy storage, and electrocatalysis. These can be exfoliated delaminated to create high surface areas interlayers accessibility. Intercalation is known possible, it critical many including electrochemical water purification, sensing. However, little about the nature intercalant bonding interactions between within MXene. We have investigated urea...
Neutron scattering measurements on the pyrochlore magnet Ce2Zr2O7 reveal an unusual crystal field splitting of its lowest J=5/2 multiplet, such that ground-state doublet is composed mJ=±3/2, giving these doublets a dipole-octupole (DO) character with local Ising anisotropy. Its magnetic susceptibility shows weak antiferromagnetic correlations θCW=−0.4(2) K, leading to naive expectation all-in, all-out ordered state at low temperatures. Instead, our low-energy inelastic neutron show dynamic...
There is widespread interest in determining the structural features of redox-active electrochemical energy storage materials that enable simultaneous high power and density. Here, we present discovery confined interlayer water crystalline tungsten oxide hydrates, WO3·nH2O, enables highly reversible proton intercalation at subsecond time scales. By comparing transformation kinetics dynamics hydrates with anhydrous WO3, determine rapid due to ability layers isolate transformations two...
Abstract The broken symmetry in the atomic-scale ordering of glassy versus crystalline solids leads to a daunting challenge provide suitable metrics for describing order within disorder, especially on length scales beyond nearest neighbor that are characterized by rich structural complexity. Here, we address this silica, canonical network-forming glass, using hot cold compression (i) systematically increase after densification and (ii) prepare two glasses with same high-density but...
We use neutron scattering to show that ferromagnetic (FM) phase transition in the two-dimensional (2D) honeycomb lattice ${\mathrm{CrI}}_{3}$ is a weakly first order and controlled by spin-orbit coupling (SOC) induced magnetic anisotropy, instead of exchange as conventional ferromagnet. With increasing temperature, magnitude seen spin gap at Brillouin zone center, decreases power law fashion vanishes ${T}_{C}$, while in-plane $c$-axis spin-wave stiffnesses associated with couplings remain...
The search for topological spin excitations in recently discovered two-dimensional (2D) van der Waals (vdW) magnetic materials is important because of their potential applications dissipationless spintronics. In the 2D vdW ferromagnetic (FM) honeycomb lattice CrI3 (TC=61 K), acoustic and optical waves are found to be separated by a gap at Dirac points. presence such signature if it arises from next-nearest-neighbor (NNN) Dzyaloshinskii-Moriya (DM) or bond-angle-dependent Kitaev interactions...
Abstract CrSBr is an air‐stable two‐dimensional (2D) van der Waals semiconducting magnet with great technological promise, but its atomic‐scale magnetic interactions—crucial information for high‐frequency switching—are poorly understood. An experimental study presented to determine the exchange Hamiltonian and bulk magnon spectrum. The A ‐type antiferromagnetic order using single crystal neutron diffraction confirmed here. dispersions are also measured inelastic scattering rigorously fit...
It is commonly accepted that the Vision Transformer model requires sophisticated regularization techniques to excel at ImageNet-1k scale data. Surprisingly, we find this not case and standard data augmentation sufficient. This note presents a few minor modifications original (ViT) vanilla training setting dramatically improve performance of plain ViT models. Notably, 90 epochs surpass 76% top-1 accuracy in under seven hours on TPUv3-8, similar classic ResNet50 baseline, 300 reach 80% less...
Vision Transformers convert images to sequences by slicing them into patches. The size of these patches controls a speed/accuracy tradeoff, with smaller leading higher accuracy at greater computational cost, but changing the patch typically requires retraining model. In this paper, we demonstrate that simply randomizing training time leads single set weights performs well across wide range sizes, making it possible tailor model different compute budgets deployment time. We extensively...