arXiv subjects

Yuhui Zhang

Publications and source records attributed to Yuhui Zhang.

89 records · Page 5Linked to original sources

arXiv2023

DGP-Net: Dense Graph Prototype Network for Few-Shot SAR Target Recognition

The inevitable feature deviation of synthetic aperture radar (SAR) image due to the special imaging principle (depression angle variation) leads to poor recognition accuracy, especially in few-shot learning (FSL). To deal with this problem, we propose a dense graph prototype network (DGP-Net) to eliminate the feature deviation by learning potential features, and classify by learning feature distribution. The role of the prototype in this model is to solve the problem of large distance between congeneric samples taken due to the contingency of single sampling in FSL, and enhance the robustness of the model. Experimental results on the MSTAR dataset show that the DGP-Net has good classification results for SAR images with different depression angles and the recognition accuracy of it is higher than typical FSL methods.

cs.CV↗

arXiv2023

Diagnosing and Rectifying Vision Models using Language

Recent multi-modal contrastive learning models have demonstrated the ability to learn an embedding space suitable for building strong vision classifiers, by leveraging the rich information in large-scale image-caption datasets. Our work highlights a distinct advantage of this multi-modal embedding space: the ability to diagnose vision classifiers through natural language. The traditional process of diagnosing model behaviors in deployment settings involves labor-intensive data acquisition and annotation. Our proposed method can discover high-error data slices, identify influential attributes and further rectify undesirable model behaviors, without requiring any visual data. Through a combination of theoretical explanation and empirical verification, we present conditions under which classifiers trained on embeddings from one modality can be equivalently applied to embeddings from another modality. On a range of image datasets with known error slices, we demonstrate that our method can effectively identify the error slices and influential attributes, and can further use language to rectify failure modes of the classifier.

cs.LG↗

arXiv2023

Adapting Pre-trained Vision Transformers from 2D to 3D through Weight Inflation Improves Medical Image Segmentation

Given the prevalence of 3D medical imaging technologies such as MRI and CT that are widely used in diagnosing and treating diverse diseases, 3D segmentation is one of the fundamental tasks of medical image analysis. Recently, Transformer-based models have started to achieve state-of-the-art performances across many vision tasks, through pre-training on large-scale natural image benchmark datasets. While works on medical image analysis have also begun to explore Transformer-based models, there is currently no optimal strategy to effectively leverage pre-trained Transformers, primarily due to the difference in dimensionality between 2D natural images and 3D medical images. Existing solutions either split 3D images into 2D slices and predict each slice independently, thereby losing crucial depth-wise information, or modify the Transformer architecture to support 3D inputs without leveraging pre-trained weights. In this work, we use a simple yet effective weight inflation strategy to adapt pre-trained Transformers from 2D to 3D, retaining the benefit of both transfer learning and depth information. We further investigate the effectiveness of transfer from different pre-training sources and objectives. Our approach achieves state-of-the-art performances across a broad range of 3D medical image datasets, and can become a standard strategy easily utilized by all work on Transformer-based models for 3D medical images, to maximize performance.

cs.CV↗

arXiv2022

Mind the Gap: Understanding the Modality Gap in Multi-modal Contrastive Representation Learning

We present modality gap, an intriguing geometric phenomenon of the representation space of multi-modal models. Specifically, we show that different data modalities (e.g. images and text) are embedded at arm's length in their shared representation in multi-modal models such as CLIP. Our systematic analysis demonstrates that this gap is caused by a combination of model initialization and contrastive learning optimization. In model initialization, we show empirically and theoretically that the representation of a common deep neural network is restricted to a narrow cone. As a consequence, in a multi-modal model with two encoders, the representations of the two modalities are clearly apart when the model is initialized. During optimization, contrastive learning keeps the different modalities separate by a certain distance, which is influenced by the temperature parameter in the loss function. Our experiments further demonstrate that varying the modality gap distance has a significant impact in improving the model's downstream zero-shot classification performance and fairness. Our code and data are available at https://modalitygap.readthedocs.io/

cs.CL↗

arXiv2022

THz detection and amplification using plasmonic Field Effect Transistors driven by DC drain currents

We report on the numerical and theoretical results of sub-THz and THz detection by a current-driven InGaAs/GaAs plasmonic Field-Effect Transistor (TeraFET). New equations are developed to account for the channel length dependence of the drain voltage and saturation current. Numerical simulation results demonstrate that the effect of drain bias current on the source-to-drain response voltage (dU) varies with the device channel length. In a long-channel TeraFET where plasmonic oscillations cannot reach the drain, dU is always positive and rises rapidly with increasing drain current. For a short device in which plasmonic oscillations reach the drain, the current-induced nonuniform electric field leads to a negative response, agreeing with previous observations. At negative dU, the amplitude of the small-signal voltage at the drain side becomes larger than that at the source side. Thus, the device effectively serves as a THz amplifier in this condition. Under the resonant mode, the negative response can be further amplified near the resonant peaks. A new expression of dU is proposed to account for this resonant effect. Based on those expressions, a current-driven TeraFET spectrometer is proposed. The ease of implementation and simplified calibration procedures make it competitive or superior compared with other TeraFET-based spectrometers.

physics.app-ph↗

arXiv2021

TeraFET terahertz detectors with spatially non-uniform gate capacitances

A non-uniform capacitance profile in the channel of a THz field-effect transistor (TeraFET) could significantly improve the THz detection performance. The analytical solutions and simulations of the hydrodynamic equations for the exponentially varying capacitance versus distance showed ~10% increase in the responsivity for the 130 nm Si TeraFETs in good agreement with numerical simulations. Using the numerical solutions of the hydrodynamic equations, we compared three different Cg configurations (exponential, linear and sawtooth). The simulations showed that the sawtooth configuration provides the largest response tunability. We also compared the effects of the non-uniform capacitance profiles for Si, III-V, and p-diamond TeraFETs. The results confirmed a great potential of p-diamond for THz applications. Varying the threshold voltage across the channel could have an effect similar to that of varying the gate-to-channel capacitance. The physics behind the demonstrated improvement in THz detection performance is related to breaking the channel symmetry by device geometry of composition asymmetry.

physics.app-ph↗

arXiv2021

Collision dominated, ballistic, and viscous regimes of terahertz plasmonic detection by graphene

The terahertz detection performance and operating regimes of graphene plasmonic field-effect transistors (FETs) were investigated by a hydrodynamic model. Continuous wave detection simulations showed that the graphene response sensitivity is similar to that of other materials including Si, InGaAs, GaN, and diamond-based FETs. However, the pulse detection results indicated a very short response time, which favors the rapid/high-sensitively detection. The analysis on the mobility dependence of the response time revealed the same detection regimes as the traditional semiconductor materials, i.e. the non-resonant (collision dominated) regime, the resonant ballistic regime, and the viscous regime. When the kinematic viscosity (ν) is above a certain critical viscosity value, νNR, the plasmonic FETs always operates in the viscous non-resonant regime regardless of channel length (L). In this regime, the response time rises monotonically with the increase of L. When ν < νNR, the plasmonic resonance can be reached in a certain range of L (i.e. the resonant window). Within this window, the carrier transport is ballistic. For a sufficiently short channel, the graphene devices would always operate in the non-resonant regime regardless of the field-effect mobility, corresponding to another viscous regime. The above work mapped the operating regimes of graphene plasmonic FETs, and demonstrated the significance of the viscous effects for the graphene plasmonic detection. These results could be used for the extraction of the temperature dependences of viscosity in graphene.

physics.app-ph↗

arXiv2020

Ultrashort Pulse Detection and Response Time Analysis Using Plasma-wave Terahertz Field Effect Transistors

We report on the response characteristics of plasmonic terahertz field-effect transistors (TeraFETs) fed with femtosecond and picosecond pulses. Varying the pulse width (tpw) from 10-15 s to 10-10 s under a constant input power condition revealed two distinctive pulse detection modes. In the short pulse mode (tpw << L/s, where L is the gated channel length, s is the plasma velocity), the source-to-drain voltage response is a sharp pulse oscillatory decay preceded by a delay time on the order of L/s. The plasma wave travels along the channel like the shallow water wave with a relatively narrow wave package. In the long pulse mode (tpw > L/s), the response profile has two oscillatory decay processes and the propagation of plasma wave is analogues to oscillating rod with one side fixed. The ultimate response time at the long pulse mode is significantly higher than that under the short pulse conditions. The detection conditions under the long pulse mode are close to the step response condition, and the response time conforms well to the analytical theory for the step function response. The simulated waveform agrees well with the measured pulse response. Our results show that the measurements of the pulse response enable the material parameter extraction from the pulse response data (including the effective mass, kinematic viscosity and momentum relaxation time).

physics.app-ph↗

arXiv2020

Biomedical and Clinical English Model Packages in the Stanza Python NLP Library

We introduce biomedical and clinical English model packages for the Stanza Python NLP library. These packages offer accurate syntactic analysis and named entity recognition capabilities for biomedical and clinical text, by combining Stanza's fully neural architecture with a wide variety of open datasets as well as large-scale unsupervised biomedical and clinical text data. We show via extensive experiments that our packages achieve syntactic analysis and named entity recognition performance that is on par with or surpasses state-of-the-art results. We further show that these models do not compromise speed compared to existing toolkits when GPU acceleration is available, and are made easy to download and use with Stanza's Python interface. A demonstration of our packages is available at: http://stanza.run/bio.

cs.CL↗

arXiv2020

p-Diamond, Si, GaN and InGaAs TeraFETs

p-diamond field effect transistors (FETs) featuring large effective mass, long momentum relaxation time and high carrier mobility are a superb candidate for plasmonic terahertz (THz) applications. Previous studies have shown that p-diamond plasmonic THz FETs (TeraFETs) could operate in plasmonic resonant mode at a low frequency window of 200 GHz to ~600 GHz, thus showing promising potential for beyond 5G sub-THz applications. In this work, we explore the advantages of p-diamond transistors over n-diamond, Si, GaN and InGaAs TeraFETs and estimate the minimum mobility required for the resonant plasmons. Our numerical simulation shows that the p-diamond TeraFET has a relatively low minimum resonant mobility, and thus could enable resonant detection. The diamond response characteristics can be adjusted by changing operating temperature. A decrease of temperature from 300 K to 77 K improves the detection performance of TeraFETs. At both room temperature and 77 K, the p-diamond TeraFET presents a high detection sensitivity in a large dynamic range. When the channel length is reduced to 20 nm, the p-diamond TeraFET exhibits the highest DC response among all types of TeraFETs in a large frequency window.

physics.app-ph↗

arXiv2020

Stanza: A Python Natural Language Processing Toolkit for Many Human Languages

We introduce Stanza, an open-source Python natural language processing toolkit supporting 66 human languages. Compared to existing widely used toolkits, Stanza features a language-agnostic fully neural pipeline for text analysis, including tokenization, multi-word token expansion, lemmatization, part-of-speech and morphological feature tagging, dependency parsing, and named entity recognition. We have trained Stanza on a total of 112 datasets, including the Universal Dependencies treebanks and other multilingual corpora, and show that the same neural architecture generalizes well and achieves competitive performance on all languages tested. Additionally, Stanza includes a native Python interface to the widely used Java Stanford CoreNLP software, which further extends its functionality to cover other tasks such as coreference resolution and relation extraction. Source code, documentation, and pretrained models for 66 languages are available at https://stanfordnlp.github.io/stanza.

cs.CL↗

arXiv2018

Large-scale Generative Modeling to Improve Automated Veterinary Disease Coding

Supervised learning is limited both by the quantity and quality of the labeled data. In the field of medical record tagging, writing styles between hospitals vary drastically. The knowledge learned from one hospital might not transfer well to another. This problem is amplified in veterinary medicine domain because veterinary clinics rarely apply medical codes to their records. We proposed and trained the first large-scale generative modeling algorithm in automated disease coding. We demonstrate that generative modeling can learn discriminative features when additionally trained with supervised fine-tuning. We systematically ablate and evaluate the effect of generative modeling on the final system's performance. We compare the performance of our model with several baselines in a challenging cross-hospital setting with substantial domain shift. We outperform competitive baselines by a large margin. In addition, we provide interpretation for what is learned by our model.

cs.CL↗

arXiv2015

Entanglement Area Law in Disordered Free Fermion Anderson Model in One, Two, and Three Dimensions

We calculate numerically the entanglement entropy of free fermion ground states in one-, two- and three-dimensional Anderson models, and find that it obeys the area law as long as the linear size of the subsystem is sufficiently larger than the mean free path. This result holds in the metallic phase of the three-dimensional Anderson model, where the mean free path is finite although the localization length is infinite. Relation between the present results and earlier ones on area law violation in special one-dimensional models that support metallic phases is discussed.

cond-mat.str-el↗

arXiv2014

Existence of strong-pairing quantum Hall phase in bilayer cold atom systems with dipolar interactions

We study bilayer fermionic cold atom systems with dipolar interactions, as well as a two-component tunable pseudopotential (TCTP) model which keeps only the zeroth and first Haldane pseudopotentials, at total Landau level filling factor 1/2. Our numerical results on the TCTP model indicates that Haldane-Rezayi state describes the critical point between strong and weak d-wave pairing quantum Hall phases. Further increasing the attractive zeroth pseudopotentials, the system transits from the strong-pairing phase to a stripe phase, and then to a cluster phase (or phase separation). The dipolar interaction can be mapped onto the TCTP model in the strong-pairing phase, if high order pseudopotentials are ignored. Our numerical results show that this is indeed the case, so the strong-pairing phase can be realized in the cold atom system.

cond-mat.str-el↗

arXiv2013

Edge reconstruction of fractional quantum Hall liquids with spin degrees of freedom

We study the interplay of confining potential, electron-electron interaction, and Zeeman splitting at the edges of fractional quantum Hall liquids, using numerical diagonalization of finite-size systems. The filling factors studied include 1/3, 5/2, 2/5, and 2/3. In the absence of Zeeman splitting and an edge, the first two have spin fully polarized ground states, while the latter two have singlet ground states. We find that with few exceptions, edge instabilities of these systems are triggered by softening of edge spin waves for Abelian fractional quantum Hall liquids (1/3, 2/5 and 2/3 liquids), and are triggered by softening of edge magnetoplasmon excitations for non-Abelian 5/2 liquid at the smoother confinement side. Phase diagrams are obtained in the accessible parameter spaces.

cond-mat.str-el↗

arXiv2013

Edge spin excitations and reconstructions of integer quantum Hall liquids

We study the effect of electron-electron interaction on the charge and spin structures at the edge of integer quantum Hall liquids, under three different kinds of confining potentials. Our exact diagonalization calculation for small systems indicates that the low energy excitations of ν=1 ferromagnetic state are bosonic edge spin waves. Instabilities of the ferromagnetic state with altering confinement strength result from the softening of these edge spin waves, and formation of edge spin textures. In ν\lesssim 2 regime, exact diagonalization on edge electron systems indicates that compact Hartree-Fock states with different total spin always become ground states in some regions of parameter space, and the ground states appear in between two compact states are their edge spin waves. The initial ν=2 instability is toward the compact state with total spin 1. Larger systems are studied using a microscopic trial wave functions, and some quantitative predictions on the edge instabilities for a certain type of confining potential are reached in the thermodynamic limit.

cond-mat.str-el↗

arXiv2012

Coulomb impurity under magnetic field in graphene: a semiclassical approach

We address the problem of a Coulomb impurity in graphene in the presence of a perpendicular uniform magnetic field. We show that the problem can be solved below the supercritical impurity magnitude within the WKB approximation. Without impurity the semiclassical energies correctly reproduce the Landau level spectrum. For a given Landau level the WKB energy depends on the absolute value of angular momentum in a way which is consistent with the exact diagonalization result. Below the supercritical impurity magnitude, the WKB solution can be expanded as a convergent series in powers of the effective fine structure constant. Relevance of our results to validity of the widely used Landau level projection approximation is discussed.

cond-mat.str-el↗