arXiv subjects

Ting Cao

Publications and source records attributed to Ting Cao.

At least 127 records · Page 7Linked to original sources

arXiv2023

ElasticViT: Conflict-aware Supernet Training for Deploying Fast Vision Transformer on Diverse Mobile Devices

Neural Architecture Search (NAS) has shown promising performance in the automatic design of vision transformers (ViT) exceeding 1G FLOPs. However, designing lightweight and low-latency ViT models for diverse mobile devices remains a big challenge. In this work, we propose ElasticViT, a two-stage NAS approach that trains a high-quality ViT supernet over a very large search space that supports a wide range of mobile devices, and then searches an optimal sub-network (subnet) for direct deployment. However, prior supernet training methods that rely on uniform sampling suffer from the gradient conflict issue: the sampled subnets can have vastly different model sizes (e.g., 50M vs. 2G FLOPs), leading to different optimization directions and inferior performance. To address this challenge, we propose two novel sampling techniques: complexity-aware sampling and performance-aware sampling. Complexity-aware sampling limits the FLOPs difference among the subnets sampled across adjacent training steps, while covering different-sized subnets in the search space. Performance-aware sampling further selects subnets that have good accuracy, which can reduce gradient conflicts and improve supernet quality. Our discovered models, ElasticViT models, achieve top-1 accuracy from 67.2% to 80.0% on ImageNet from 60M to 800M FLOPs without extra retraining, outperforming all prior CNNs and ViTs in terms of accuracy and latency. Our tiny and small models are also the first ViT models that surpass state-of-the-art CNNs with significantly lower latency on mobile devices. For instance, ElasticViT-S1 runs 2.62x faster than EfficientNet-B0 with 0.1% higher accuracy.

cs.CV↗

arXiv2023

SpaceEvo: Hardware-Friendly Search Space Design for Efficient INT8 Inference

The combination of Neural Architecture Search (NAS) and quantization has proven successful in automatically designing low-FLOPs INT8 quantized neural networks (QNN). However, directly applying NAS to design accurate QNN models that achieve low latency on real-world devices leads to inferior performance. In this work, we find that the poor INT8 latency is due to the quantization-unfriendly issue: the operator and configuration (e.g., channel width) choices in prior art search spaces lead to diverse quantization efficiency and can slow down the INT8 inference speed. To address this challenge, we propose SpaceEvo, an automatic method for designing a dedicated, quantization-friendly search space for each target hardware. The key idea of SpaceEvo is to automatically search hardware-preferred operators and configurations to construct the search space, guided by a metric called Q-T score to quantify how quantization-friendly a candidate search space is. We further train a quantized-for-all supernet over our discovered search space, enabling the searched models to be directly deployed without extra retraining or quantization. Our discovered models establish new SOTA INT8 quantized accuracy under various latency constraints, achieving up to 10.1% accuracy improvement on ImageNet than prior art CNNs under the same latency. Extensive experiments on diverse edge devices demonstrate that SpaceEvo consistently outperforms existing manually-designed search spaces with up to 2.5x faster speed while achieving the same accuracy.

cs.CV↗

arXiv2023

Gamify Stencil Dwarf on Cloud for Democratizing Scientific Computing

Stencil computation is one of the most important kernels in various scientific computing. Nowadays, most Stencil-driven scientific computing still relies heavily on supercomputers, suffering from expensive access, poor scalability, and duplicated optimizations. This paper proposes Tetris, the first system for high-performance Stencil on heterogeneous CPU+GPU, towards democratizing Stencil-driven scientific computing on Cloud. In Tetris, polymorphic tiling tetrominoes are first proposed to bridge different hardware architectures and various application contexts with a perfect spatial and temporal tessellation automatically. Tetris is contributed by three main components: (1) Underlying hardware characteristics are first captured to achieve a sophisticated Pattern Mapping by register-level tetrominoes; (2) An efficient Locality Enhancer is first presented for data reuse on spatial and temporal dimensions simultaneously by cache/SMEM-level tetrominoes; (3) A novel Concurrent Scheduler is first designed to exploit the full potential of on-cloud memory and computing power by memory-level tetrominoes. Tetris is orthogonal to (and complements) the optimizations or deployments for a wide variety of emerging and legacy scientific computing applications. Results of thermal diffusion simulation demonstrate that the performance is improved by 29.6x, reducing time cost from day to hour, while preserving the original accuracy.

cs.DC↗

arXiv2023

Kekulé Moiré Superlattices

Moiré superlattices from stacks of van der Waals materials offer an exciting arena in the fields of condensed matter physics and materials science. Typically, these moiré superlattices consist of materials with identical or similar structures, and the long moiré period arises from a small twist angle or lattice mismatch. In this article, we discuss that long moiré period appears in a new moiré system by stacking two dissimilar van der Waals layers with large lattice mismatch, resulting in couplings between moiré bands from remote valleys in the momentum space. In this system, the first layer is reconstructed using a $\sqrt{3}$ by $\sqrt{3}$ supercell that resembles the Kekulé distortion in graphene, and such reconstruction becomes nearly commensurate with the second layer. This Kekulé moiré superlattice is realized in heterostructures of transition metal dichalcogenides and metal phosphorus trichalcogenides such as MoTe$_2$/MnPSe$_3$. By first-principles calculations, we demonstrate that the antiferromagnetic MnPSe$_3$ strongly couples the otherwise degenerate Kramers' valleys of MoTe$_2$, resulting in valley pseudospin textures that depend on Néel vector direction, stacking geometry, and external fields. With one hole per moiré supercell, we predict that the system can become a Chern insulator, of which the topology is tunable by external fields.

cond-mat.mes-hall↗

arXiv2023

Tunable phononic coupling in excitonic quantum emitters

Engineering the coupling between fundamental quantum excitations is at the heart of quantum science and technologies. A significant case is the creation of quantum light sources in which coupling between single photons and phonons can be controlled and harnessed to enable quantum information transduction. Here, we report the deterministic creation of quantum emitters featuring highly tunable coupling between excitons and phonons. The quantum emitters are formed in strain-induced quantum dots created in homobilayer semiconductor WSe2. The colocalization of quantum confined interlayer excitons and THz interlayer breathing mode phonons, which directly modulate the exciton energy, leads to a uniquely strong phonon coupling to single-photon emission. The single-photon spectrum of interlayer exciton emission features a single-photon purity >83% and multiple phonon replicas, each heralding the creation of a phonon Fock state in the quantum emitter. Owing to the vertical dipole moment of the interlayer exciton, the phonon-photon interaction is electrically tunable in a wide range, promising to reach the strong coupling regime. Our result demonstrates a new type of solid-state quantum excitonic-optomechanical system at the atomic interface that emits flying photonic qubits coupled with stationary phonons, which could be exploited for quantum transduction and interconnection.

cond-mat.mes-hall↗

arXiv2023

Strain-programmable van der Waals magnetic tunnel junctions

The magnetic tunnel junction (MTJ) is a backbone device for spintronics. Realizing next generation energy efficient MTJs will require operating mechanisms beyond the standard means of applying magnetic fields or large electrical currents. Here, we demonstrate a new concept for programmable MTJ operation via strain control of the magnetic states of CrSBr, a layered antiferromagnetic semiconductor used as the tunnel barrier. Switching the CrSBr from antiferromagnetic to ferromagnetic order generates a giant tunneling magnetoresistance ratio without external magnetic field at temperatures up to ~ 140 K. When the static strain is set near the phase transition, applying small strain pulses leads to active flipping of layer magnetization with controlled layer number and thus magnetoresistance states. Further, finely adjusting the static strain to a critical value turns on stochastic switching between metastable states, with a strain-tunable sigmoidal response curve akin to the stochastic binary neuron. Our results highlight the potential of strain-programmable van der Waals MTJs towards spintronic applications, such as magnetic memory, random number generation, and probabilistic and neuromorphic computing.

cond-mat.mes-hall↗

arXiv2022

Spectroscopic evidence of flat bands in breathing kagome semiconductor Nb3I8

Kagome materials have become solid grounds to study the interplay among geometry, topology, correlation, and magnetism. Recently, semiconductors Nb3X8(X = Cl, Br, I) have been predicted to be two-dimensional (2D) magnets and importantly these materials possess breathing kagome geometry. Electronic structure study of these promising materials is still lacking. Here, we report the spectroscopic evidence of at and weakly dispersing bands in breathing-kagome semiconductor Nb3I8 around 500 meV binding energy, which is well supported by our first-principles calculations. These bands originate from the breathing kagome lattice of Niobium atoms and have Nb d character. They are found to be sensitive to polarization of the incident photon beam. Our study provides insight into the electronic structure and at band topology in an exfoliable kagome semiconductor thereby providing an important platform to understand the interaction of geometry and electron correlations in 2D material.

cond-mat.mes-hall↗

arXiv2022

Mixed-dimensional moiré systems of graphitic thin films with a twisted interface

Moiré patterns formed by stacking atomically-thin van der Waals crystals with a relative twist angle can give rise to dramatic new physical properties. The study of moiré materials has so far been limited to structures comprising no more than a few vdW sheets, since a moiré pattern localized to a single two-dimensional interface is generally assumed to be incapable of appreciably modifying the properties of a bulk three-dimensional crystal. Layered semimetals such as graphite offer a unique platform to challenge this paradigm, owing to distinctive properties arising from their nearly-compensated electron and hole bulk doping. Here, we perform transport measurements of dual-gated devices constructed by slightly rotating a monolayer graphene sheet atop a thin bulk graphite crystal. We find that the moiré potential transforms the electronic properties of the entire bulk graphitic thin film. At zero and small magnetic fields, transport is mediated by a combination of gate-tunable moiré and graphite surface states, as well as coexisting semimetallic bulk states that do not respond to gating. At high field, the moiré potential hybridizes with the graphitic bulk states owing to the unique properties of the two lowest Landau bands of graphite. These Landau bands facilitate the formation of a single quasi-two-dimensional hybrid structure in which the moiré and bulk graphite states are inextricably mixed. Our results establish twisted graphene-graphite as the first in a new class of mixed-dimensional moiré materials.

cond-mat.mes-hall↗

arXiv2022

Designing magnetic properties in CrSBr through hydrostatic pressure and ligand substitution

The ability to control magnetic properties of materials is crucial for fundamental research and underpins many information technologies. In this context, two-dimensional materials are a particularly exciting platform due to their high degree of tunability and ease of implementation into nanoscale devices. Here we report two approaches for manipulating the A-type antiferromagnetic properties of the layered semiconductor CrSBr through hydrostatic pressure and ligand substitution. Hydrostatic pressure compresses the unit cell, increasing the interlayer exchange energy while lowering the Néel temperature. Ligand substitution, realized synthetically through Cl alloying, anisotropically compresses the unit cell and suppresses the Cr-halogen covalency, reducing the magnetocrystalline anisotropy energy and decreasing the Néel temperature. A detailed structural analysis combined with first-principles calculations reveal that alterations in the magnetic properties are intricately related to changes in direct Cr-Cr exchange interactions and the Cr-anion superexchange pathways. Further, we demonstrate that Cl alloying enables chemical tuning of the interlayer coupling from antiferromagnetic to ferromagnetic, which is unique amongst known two-dimensional magnets. The magnetic tunability, combined with a high ordering temperature, chemical stability, and functional semiconducting properties, make CrSBr an ideal candidate for pre- and post-synthetic design of magnetism in two-dimensional materials.

cond-mat.mtrl-sci↗

arXiv2022

Tunable Exciton-Hybridized Magnon Interactions in a Layered Semiconductor

The interaction between distinct excitations in solids is of both fundamental interest and technological importance. One example of such interactions is coupling between an exciton, a Coulomb bound electron-hole pair, and a magnon, a collective spin excitation. The recent emergence of van der Waals magnetic semiconductors provides a powerful platform for exploring these exciton-magnon interactions and their fundamental properties, such as strong correlation, as well as their photo-spintronic and quantum transduction applications. Here we demonstrate precise control of coherent exciton-magnon interactions in the layered magnetic semiconductor CrSBr. We show that by controlling the direction of applied magnetic fields relative to the crystal axes, and thus the rotational symmetry of the magnetic system, we can tune not only the exciton coupling to the bright magnon, but also to an optically dark mode via magnon hybridization. The exciton-magnon coupling and associated magnon dispersion curves can be further modulated by applying a uniaxial strain. At the critical strain, a dispersionless dark magnon band emerges. Our results demonstrate unprecedented control of the opto-mechanical-magnonic coupling, and a step towards the predictable and controllable implementation of hybrid quantum magnonics.

cond-mat.mes-hall↗

arXiv2022

SwiftPruner: Reinforced Evolutionary Pruning for Efficient Ad Relevance

Ad relevance modeling plays a critical role in online advertising systems including Microsoft Bing. To leverage powerful transformers like BERT in this low-latency setting, many existing approaches perform ad-side computations offline. While efficient, these approaches are unable to serve cold start ads, resulting in poor relevance predictions for such ads. This work aims to design a new, low-latency BERT via structured pruning to empower real-time online inference for cold start ads relevance on a CPU platform. Our challenge is that previous methods typically prune all layers of the transformer to a high, uniform sparsity, thereby producing models which cannot achieve satisfactory inference speed with an acceptable accuracy. In this paper, we propose SwiftPruner - an efficient framework that leverages evolution-based search to automatically find the best-performing layer-wise sparse BERT model under the desired latency constraint. Different from existing evolution algorithms that conduct random mutations, we propose a reinforced mutator with a latency-aware multi-objective reward to conduct better mutations for efficiently searching the large space of layer-wise sparse models. Extensive experiments demonstrate that our method consistently achieves higher ROC AUC and lower latency than the uniform sparse baseline and state-of-the-art search methods. Remarkably, under our latency requirement of 1900us on CPU, SwiftPruner achieves a 0.86% higher AUC than the state-of-the-art uniform sparse baseline for BERT-Mini on a large scale real-world dataset. Online A/B testing shows that our model also achieves a significant 11.7% cut in the ratio of defective cold start ads with satisfactory real-time serving latency.

cs.IR↗

arXiv2022

Turbo: Opportunistic Enhancement for Edge Video Analytics

Edge computing is being widely used for video analytics. To alleviate the inherent tension between accuracy and cost, various video analytics pipelines have been proposed to optimize the usage of GPU on edge nodes. Nonetheless, we find that GPU compute resources provisioned for edge nodes are commonly under-utilized due to video content variations, subsampling and filtering at different places of a pipeline. As opposed to model and pipeline optimization, in this work, we study the problem of opportunistic data enhancement using the non-deterministic and fragmented idle GPU resources. In specific, we propose a task-specific discrimination and enhancement module and a model-aware adversarial training mechanism, providing a way to identify and transform low-quality images that are specific to a video pipeline in an accurate and efficient manner. A multi-exit model structure and a resource-aware scheduler is further developed to make online enhancement decisions and fine-grained inference execution under latency and GPU resource constraints. Experiments across multiple video analytics pipelines and datasets reveal that by judiciously allocating a small amount of idle resources on frames that tend to yield greater marginal benefits from enhancement, our system boosts DNN object detection accuracy by $7.3-11.3\%$ without incurring any latency costs.

cs.CV↗

arXiv2022

Intercell Moiré Exciton Complexes in Electron Lattices

Excitons, Coulomb-bound electron-hole pairs, play a fundamental role in both optical excitation and correlated phenomena in solids. When an exciton interacts with other quasi-particles, few- and many-body excited states, such as trions, exciton Fermi-polarons, Mahan excitons can appear. Here, we report a new interaction between exciton and charges enabled by unusual quantum confinement in 2D moiré superlattices, which results in novel exciton many-body ground states composed of moiré excitons and correlated electron lattices. Unique to H-stacked (or 60o-twisted) WS2/WSe2 heterobilayer, we found that the interlayer atomic registry and moiré structural reconstruction leads to an interlayer moiré exciton (IME) whose hole in one layer is surrounded by its partner electron's wavefunction spread among three adjacent moiré traps in the other layer. This 3D excitonic structure can enable large in-plane electrical quadrupole moments in addition to the vertical dipole. Upon doping, the electric quadrupole facilitates the binding of IME to the charges in neighboring moiré cells, forming an intercell charged exciton complex. The exciton complex is unveiled by the IME photoluminescence energy jumps when the electron lattices form at both fractional and integer-filled moiré minibands, with replica-like spectral features between successive integer moiré fillings. Our work provides the framework in understanding and engineering emergent exciton many-body states in correlated moiré charge orders.

cond-mat.mes-hall↗

arXiv2022

Long-range transport of 2D excitons with acoustic waves

Excitons are elementary optical excitation in semiconductors. The ability to manipulate and transport these quasiparticles would enable excitonic circuits and devices for quantum photonic technologies. Recently, interlayer excitons in 2D semiconductors have emerged as a promising candidate for engineering excitonic devices due to their long lifetime, large exciton binding energy, and gate tunability. However, the charge-neutral nature of the excitons leads to weak response to the in-plane electric field and thus inhibits transport beyond the diffusion length. Here, we demonstrate the directional transport of interlayer excitons in bilayer WSe2 driven by the propagating potential traps induced by surface acoustic waves (SAW). We show that at 100 K, the SAW-driven excitonic transport is activated above a threshold acoustic power and reaches 20 mm, a distance at least ten times longer than the diffusion length and only limited by the device size. Temperature-dependent measurement reveals the transition from the diffusion-limited regime at low temperature to the acoustic field-driven regime at elevated temperature. Our work shows that acoustic waves are an effective, contact-free means to control exciton dynamics and transport, promising for realizing 2D materials-based excitonic devices such as exciton transistors, switches, and transducers up to room temperature.

cond-mat.mes-hall↗

arXiv2022

Ab initio calculations of spin-nonconserving exciton-phonon scattering in monolayer transition metal dichalcogenides

We investigate the spin-nonconserving relaxation channel of excitons by their couplings with phonons in two-dimensional transition metal dichalcogenides using $\textit{ab initio}$ approaches. Combining $\text{GW}$-Bethe-Salpeter equation method and density functional perturbation theory, we calculate the electron-phonon and exciton-phonon coupling matrix elements for the spin-flip scattering in monolayer WSe$_{\text{2}}$, and further analyze the microscopic mechanisms influencing these scattering strengths. We find that phonons could produce effective in-plane magnetic fields which flip spin of excitons, giving rise to relaxation channels complimentary to the spin-conserving relaxation. Finally, we calculate temperature-dependent spin-flip exciton-phonon relaxation times. Our method and analysis can be generalized to study other two-dimensional materials and would stimulate experimental measurements of spin-flip exciton relaxation dynamics.

cond-mat.mtrl-sci↗

arXiv2022

In-Bed Human Pose Estimation from Unseen and Privacy-Preserving Image Domains

Medical applications have benefited greatly from the rapid advancement in computer vision. Considering patient monitoring in particular, in-bed human posture estimation offers important health-related metrics with potential value in medical condition assessments. Despite great progress in this domain, it remains challenging due to substantial ambiguity during occlusions, and the lack of large corpora of manually labeled data for model training, particularly with domains such as thermal infrared imaging which are privacy-preserving, and thus of great interest. Motivated by the effectiveness of self-supervised methods in learning features directly from data, we propose a multi-modal conditional variational autoencoder (MC-VAE) capable of reconstructing features from missing modalities seen during training. This approach is used with HRNet to enable single modality inference for in-bed pose estimation. Through extensive evaluations, we demonstrate that body positions can be effectively recognized from the available modality, achieving on par results with baseline models that are highly dependent on having access to multiple modes at inference time. The proposed framework supports future research towards self-supervised learning that generates a robust model from a single source, and expects it to generalize over many unknown distributions in clinical environments.

cs.CV↗

arXiv2021

Phonon-mediated exciton relaxation in two-dimensional semiconductors: selection rules and relaxation pathways

Exciton-phonon coupling (ExPC) is crucial for energy relaxation in semiconductors, yet the first-principles calculation of such coupling remains challenging, especially for low-dimensional systems. Here, an accurate algorithm for calculating ExPC is developed and applied in exciton relaxation problems in monolayer WSe2. Considering the interplay between the exciton wave functions and electron-phonon coupling (EPC) matrix elements, we find that ExPC shows distinct selection rules from the ones of EPC. By employing the Wannier exciton model, we generalize these selection rules, which state that the angular quantum numbers of the exciton must match the winding numbers of the EPC matrix elements for the ExPC to be allowed. To verify our theory and algorithm, we calculate inter-valley exciton relaxation pathways, which agrees well with a recent experiment.

cond-mat.mtrl-sci↗

arXiv2021

Interlayer Electronic Coupling on Demand in a 2D Magnetic Semiconductor

When monolayers of two-dimensional (2D) materials are stacked into van der Waals structures, interlayer electronic coupling can introduce entirely new properties, as exemplified by recent discoveries of moiré bands that host highly correlated electronic states and quantum dot-like interlayer exciton lattices. Here we show the magnetic control of interlayer electronic coupling, as manifested in tunable excitonic transitions, in an A-type antiferromagnetic 2D semiconductor CrSBr. Excitonic transitions in bilayer and above can be drastically changed when the magnetic order is switched from layered antiferromagnetic to the field-induced ferromagnetic state, an effect attributed to the spin-allowed interlayer hybridization of electron and hole orbitals in the latter, as revealed by GW-BSE calculations. Our work uncovers a magnetic approach to engineer electronic and excitonic effects in layered magnetic semiconductors.

cond-mat.mes-hall↗