arXiv subjects

Hua Xu

Publications and source records attributed to Hua Xu.

At least 127 records · Page 7Linked to original sources

arXiv2021

Deciphering the Intense Post-Gap Absorptions of Monolayer Transition Metal Dichalcogenides

Rich valleytronics and diverse defect-induced or interlayer pre-bandgap excitonics have been extensively studied in transition metal dichalcogenides (TMDCs), a system with fascinating optical physics. However, more intense high-energy absorption peaks (~ 3 eV) above the bandgaps used to be long ignored and their underlying physical origin remains to be unveiled. Here, we employ momentum resolved electron energy loss spectroscopy to measure the dispersive behaviors of the valley excitons and intense higher-energy peaks at finite momenta. Combined with accurate Bethe Salpeter equation calculations, non-band-nesting transitions at Q valley and at midpoint of KM are found to be responsible for the high-energy broad absorption peaks in tungsten dichalcogenides and present spin polarizations similar to A excitons, in contrast with the band-nesting mechanism in molybdenum dichalcogenides. Our experiment-theory joint research will offer insights into the physical origins and manipulation of the intense high-energy excitons in TMDCs-based optoelectronic devices.

cond-mat.mtrl-sci↗

arXiv2021

Deep Open Intent Classification with Adaptive Decision Boundary

Open intent classification is a challenging task in dialogue systems. On the one hand, it should ensure the quality of known intent identification. On the other hand, it needs to detect the open (unknown) intent without prior knowledge. Current models are limited in finding the appropriate decision boundary to balance the performances of both known intents and the open intent. In this paper, we propose a post-processing method to learn the adaptive decision boundary (ADB) for open intent classification. We first utilize the labeled known intent samples to pre-train the model. Then, we automatically learn the adaptive spherical decision boundary for each known class with the aid of well-trained features. Specifically, we propose a new loss function to balance both the empirical risk and the open space risk. Our method does not need open intent samples and is free from modifying the model architecture. Moreover, our approach is surprisingly insensitive with less labeled data and fewer known intents. Extensive experiments on three benchmark datasets show that our method yields significant improvements compared with the state-of-the-art methods. The codes are released at https://github.com/thuiar/Adaptive-Decision-Boundary.

cs.CL↗

arXiv2021

Discovering New Intents with Deep Aligned Clustering

Discovering new intents is a crucial task in dialogue systems. Most existing methods are limited in transferring the prior knowledge from known intents to new intents. They also have difficulties in providing high-quality supervised signals to learn clustering-friendly features for grouping unlabeled intents. In this work, we propose an effective method, Deep Aligned Clustering, to discover new intents with the aid of the limited known intent data. Firstly, we leverage a few labeled known intent samples as prior knowledge to pre-train the model. Then, we perform k-means to produce cluster assignments as pseudo-labels. Moreover, we propose an alignment strategy to tackle the label inconsistency problem during clustering assignments. Finally, we learn the intent representations under the supervision of the aligned pseudo-labels. With an unknown number of new intents, we predict the number of intent categories by eliminating low-confidence intent-wise clusters. Extensive experiments on two benchmark datasets show that our method is more robust and achieves substantial improvements over the state-of-the-art methods. The codes are released at https://github.com/thuiar/DeepAligned-Clustering.

cs.CL↗

arXiv2021

Learning Modality-Specific Representations with Self-Supervised Multi-Task Learning for Multimodal Sentiment Analysis

Representation Learning is a significant and challenging task in multimodal learning. Effective modality representations should contain two parts of characteristics: the consistency and the difference. Due to the unified multimodal annotation, existing methods are restricted in capturing differentiated information. However, additional uni-modal annotations are high time- and labor-cost. In this paper, we design a label generation module based on the self-supervised learning strategy to acquire independent unimodal supervisions. Then, joint training the multi-modal and uni-modal tasks to learn the consistency and difference, respectively. Moreover, during the training stage, we design a weight-adjustment strategy to balance the learning progress among different subtasks. That is to guide the subtasks to focus on samples with a larger difference between modality supervisions. Last, we conduct extensive experiments on three public multimodal baseline datasets. The experimental results validate the reliability and stability of auto-generated unimodal supervisions. On MOSI and MOSEI datasets, our method surpasses the current state-of-the-art methods. On the SIMS dataset, our method achieves comparable performance than human-annotated unimodal labels. The full codes are available at https://github.com/thuiar/Self-MM.

cs.CL↗

arXiv2020

Robustly Pre-trained Neural Model for Direct Temporal Relation Extraction

Background: Identifying relationships between clinical events and temporal expressions is a key challenge in meaningfully analyzing clinical text for use in advanced AI applications. While previous studies exist, the state-of-the-art performance has significant room for improvement. Methods: We studied several variants of BERT (Bidirectional Encoder Representations using Transformers) some involving clinical domain customization and the others involving improved architecture and/or training strategies. We evaluated these methods using a direct temporal relations dataset which is a semantically focused subset of the 2012 i2b2 temporal relations challenge dataset. Results: Our results show that RoBERTa, which employs better pre-training strategies including using 10x larger corpus, has improved overall F measure by 0.0864 absolute score (on the 1.00 scale) and thus reducing the error rate by 24% relative to the previous state-of-the-art performance achieved with an SVM (support vector machine) model. Conclusion: Modern contextual language modeling neural networks, pre-trained on a large corpus, achieve impressive performance even on highly-nuanced clinical temporal relation tasks.

cs.CL↗

arXiv2020

A Post-processing Method for Detecting Unknown Intent of Dialogue System via Pre-trained Deep Neural Network Classifier

With the maturity and popularity of dialogue systems, detecting user's unknown intent in dialogue systems has become an important task. It is also one of the most challenging tasks since we can hardly get examples, prior knowledge or the exact numbers of unknown intents. In this paper, we propose SofterMax and deep novelty detection (SMDN), a simple yet effective post-processing method for detecting unknown intent in dialogue systems based on pre-trained deep neural network classifiers. Our method can be flexibly applied on top of any classifiers trained in deep neural networks without changing the model architecture. We calibrate the confidence of the softmax outputs to compute the calibrated confidence score (i.e., SofterMax) and use it to calculate the decision boundary for unknown intent detection. Furthermore, we feed the feature representations learned by the deep neural networks into traditional novelty detection algorithm to detect unknown intents from different perspectives. Finally, we combine the methods above to perform the joint prediction. Our method classifies examples that differ from known intents as unknown and does not require any examples or prior knowledge of it. We have conducted extensive experiments on three benchmark dialogue datasets. The results show that our method can yield significant improvements compared with the state-of-the-art baselines

cs.CL↗

arXiv2019

Discovering New Intents via Constrained Deep Adaptive Clustering with Cluster Refinement

Identifying new user intents is an essential task in the dialogue system. However, it is hard to get satisfying clustering results since the definition of intents is strongly guided by prior knowledge. Existing methods incorporate prior knowledge by intensive feature engineering, which not only leads to overfitting but also makes it sensitive to the number of clusters. In this paper, we propose constrained deep adaptive clustering with cluster refinement (CDAC+), an end-to-end clustering method that can naturally incorporate pairwise constraints as prior knowledge to guide the clustering process. Moreover, we refine the clusters by forcing the model to learn from the high confidence assignments. After eliminating low confidence assignments, our approach is surprisingly insensitive to the number of clusters. Experimental results on the three benchmark datasets show that our method can yield significant improvements over strong baselines.

cs.CL↗

arXiv2019

Enhancing Clinical Concept Extraction with Contextual Embeddings

Neural network-based representations ("embeddings") have dramatically advanced natural language processing (NLP) tasks, including clinical NLP tasks such as concept extraction. Recently, however, more advanced embedding methods and representations (e.g., ELMo, BERT) have further pushed the state-of-the-art in NLP, yet there are no common best practices for how to integrate these representations into clinical tasks. The purpose of this study, then, is to explore the space of possible options in utilizing these new models for clinical concept extraction, including comparing these to traditional word embedding methods (word2vec, GloVe, fastText). Both off-the-shelf open-domain embeddings and pre-trained clinical embeddings from MIMIC-III are evaluated. We explore a battery of embedding methods consisting of traditional word embeddings and contextual embeddings, and compare these on four concept extraction corpora: i2b2 2010, i2b2 2012, SemEval 2014, and SemEval 2015. We also analyze the impact of the pre-training time of a large language model like ELMo or BERT on the extraction performance. Last, we present an intuitive way to understand the semantic information encoded by contextual embeddings. Contextual embeddings pre-trained on a large clinical corpus achieves new state-of-the-art performances across all concept extraction tasks. The best-performing model outperforms all state-of-the-art methods with respective F1-measures of 90.25, 93.18 (partial), 80.74, and 81.65. We demonstrate the potential of contextual embeddings through the state-of-the-art performance these methods achieve on clinical concept extraction. Additionally, we demonstrate contextual embeddings encode valuable semantic information not accounted for in traditional word representations.

cs.CL↗

arXiv2019

BERT-based Ranking for Biomedical Entity Normalization

Developing high-performance entity normalization algorithms that can alleviate the term variation problem is of great interest to the biomedical community. Although deep learning-based methods have been successfully applied to biomedical entity normalization, they often depend on traditional context-independent word embeddings. Bidirectional Encoder Representations from Transformers (BERT), BERT for Biomedical Text Mining (BioBERT) and BERT for Clinical Text Mining (ClinicalBERT) were recently introduced to pre-train contextualized word representation models using bidirectional Transformers, advancing the state-of-the-art for many natural language processing tasks. In this study, we proposed an entity normalization architecture by fine-tuning the pre-trained BERT / BioBERT / ClinicalBERT models and conducted extensive experiments to evaluate the effectiveness of the pre-trained models for biomedical entity normalization using three different types of datasets. Our experimental results show that the best fine-tuned models consistently outperformed previous methods and advanced the state-of-the-art for biomedical entity normalization, with up to 1.17% increase in accuracy.

cs.IR↗

arXiv2019

Global haze aerosol distribution: a direct view by Geofen-5 satellite with 3.3 km spatial resolution

The Directional Polarimetric Camera (DPC) is the first Chinese multi-angle polarized earth observation satellite sensor, which has been launched onboard the GaoFen-5 Satellite in Chinese High-resolution Earth Observation Program. GaoFen-5 runs in a sun-synchronous orbit with the 2-days revisiting period. DPC employed a charge coupled device detection unit, and can realize spatial resolution of 3.3 km under a swath width of 1850 km. Moreover, DPC has 3 polarized channels together with 5 non-polarized bands, and is able to obtain at least 9 viewing angles by continuously capturing series images over the same target on orbit. Based on the Directional Polarization Camera (DPC) onboard GF-5 satellite, the first global high-resolution (3.3 km) map of fine-mode aerosol optical depth (AODf) over land has been obtained together by Aerospace Information Institute, Chinese Academy of Sciences, Anhui Institute of Optics and Fine Mechanics, Chinese Academy of Sciences (the manufacturer of DPC sensor) and other institutes. This AODf remote sensing observation dataset has the highest spatial resolution in the world. It can reflect the spatial information of major air pollutants (PM2.5, etc.) and provide critical basic products for "decryption" of global haze distribution.

physics.ao-ph↗

arXiv2019

Study on the spectral reconstruction of typical surface types based on spectral library and principal component analysis

To meet the demanding of spectral reconstruction in the visible and near-infrared wavelength, the spectral reconstruction method for typical surface types is discussed based on the USGS /ASTER spectral library and principal component analysis (PCA). A new spectral reconstructed model is proposed by the information of several typical bands instead of all of the wavelength bands, and a linear combination spectral reconstruction model is also discussed. By selecting 4 typical spectral datasets including green vegetation, bare soil, rangeland and concrete in the spectral range of 400-900 nm, the PCA results show that 6 principal components could characterized the spectral dataset, and the relative reconstructed errors are smaller than 2%. If only 6-7 selected typical bands are employed to spectral reconstruction for all the surface reflectance in 400-900 nm, except that the reconstructed error of green vegetation is about 3.3%, the relative errors of other 3 datasets are all smaller than 1.6%. The correlation coefficients of those 4 datasets are all larger than 0.99, which can effectively satisfy the needs of spectral reconstruction. In addition, based on the spectral library and the linear combination model of 4 common used bands of satellite remote sensing such as 490, 555, 670 and 865 nm, the reconstructed errors are smaller than 8.5% in high reflectance region and smaller than 1.5% in low reflectance region respectively, which basically meet the needs of spectral reconstruction. This study can provide a reference value for the surface reflectance processing and spectral reconstruction in satellite remote sensing research.

eess.IV↗

arXiv2019

Deep Unknown Intent Detection with Margin Loss

Identifying the unknown (novel) user intents that have never appeared in the training set is a challenging task in the dialogue system. In this paper, we present a two-stage method for detecting unknown intents. We use bidirectional long short-term memory (BiLSTM) network with the margin loss as the feature extractor. With margin loss, we can learn discriminative deep features by forcing the network to maximize inter-class variance and to minimize intra-class variance. Then, we feed the feature vectors to the density-based novelty detection algorithm, local outlier factor (LOF), to detect unknown intents. Experiments on two benchmark datasets show that our method can yield consistent improvements compared with the baseline methods.

cs.CL↗

arXiv2012

On the Limiting Shape of Young Tableaux Associated With Inhomogeneous Random Words

The limiting shape of the random Young diagrams associated with an inhomogeneous random word is identified as a multidimensional Brownian functional. This functional is identical in law to the spectrum of a random matrix. The Poissonized word problem is also briefy studied, and the asymptotic behavior of the shape analyzed.

math.PR↗

arXiv2010

Active switching in metamaterials using polarization control of light

We demonstrate on-demand control of localized surface plasmons in metamaterials by means of incident light polarization. An asymmetric mode, selectively excited by s-polarized light, interfere destructively with a bright element, thereby allowing the incident light to propagate at a fairly low loss, corresponding to electromagnetically induced transparency (EIT) in an atomic system. In contrast, a symmetric mode, excited by p-polarized light, directly couples with the incident light, which is analogous to the switch-off of EIT. The light polarization-dependent excitation of asymmetric and symmetric plasmon modes holds potential for active switching applications of plasmon hybridization.

quant-ph↗

arXiv2009

Phase-sensitive Harmonic Measurements of Microwave Nonlinearities in Cuprate Thin Films

Investigations of the intrinsic electromagnetic nonlinearity of superconductors give insight into the fundamental physics of these materials. Phase-sensitive third-order harmonic voltage data $\tilde{u}_{3f}=|u_{3f}|exp(iϕ_{3f})$ are acquired with a near-field microwave microscope on homogeneous YBa$_2$Cu$_3$O$_{7-δ}$ (YBCO) thin films in a temperature range close to the critical temperature T$_c$. As temperature is increased from below T$_c$, the harmonic magnitude exhibits a maximum, while the phase, $π/2$ in the superconducting state, goes through a minimum. It is found that samples with doping ranges from near optimal ($δ=0.16$) to underdoped ($δ=0.47$) exhibit different behavior in terms of both the harmonic magnitude and phase. In optimally-doped samples, the harmonic magnitude reaches its maximum at a temperature $T_M$ slightly lower than that associated with the minimum of phase $T_m$ and drops into the noisefloor as soon as $T_m$ is exceeded. In underdoped samples $T_M$ is shifted toward lower temperatures with respect to $T_m$ and the harmonic voltage magnitude decreases slower with temperature than in the case of optimally-doped samples. A field-based analytical model of $\tilde{u}_{3f}$ is presented, where the nonlinear behavior is introduced as corrections to the low-field, linear-response complex conductivity. The model reproduces the low-temperature regime where the $σ_2$ nonlinearity dominates, in agreement with published theoretical and experimental results. Additionally the model identifies $T_m$ as the temperature where the order parameter relaxation time becomes comparable to the microwave probing period and reproduces semi-quantitatively the experimental data.

cond-mat.supr-con↗

arXiv2009

Role of Magnetic Plasmon Resonance in Plasmonic Electromagnetically-Induced Transparency

We find that the magnetic plasmon resonance plays a vital role in plasmonic electromagnetically-induced transparency (EIT) proposed by Zhang et al. [\prl \textbf{101}, 047401 (2008)] as well as the localized surface plasmon polaritons. Based on this picture, the plasmonic EIT control through single optical field is suggested in a modified scheme for active plasmonic switching by simply adjusting the incident angle of the optical field. The tunability of plasmonic EIT in this scheme is exhibited sufficiently.

cond-mat.mtrl-sci↗

arXiv2008

Dynamical scaling of $YBa_2Cu_3O_{7-δ}$ thin film conductivity in zero field

We study dynamic fluctuation effects of $YBa_2Cu_3O_{7-δ}$ thin films in zero field around $T_c$ by doing frequency-dependent microwave conductivity measurements at different powers. The length scales probed in the experiments are varied systematically allowing us to analyze data which are not affected by the finite thickness of the films, and to observe single-parameter scaling. DC current-voltage characteristics have also been measured to independently probe fluctuations in the same samples. The combination of DC and microwave measurements allows us to precisely determine critical parameters. Our results give a dynamical scaling exponent $z=1.55\pm0.15$, which is consistent with model E-dynamics.

cond-mat.supr-con↗

arXiv2008

Frequency- and electric-field-dependent conductivity of single-walled carbon nanotube networks of varying density

We present measurements of the frequency and electric field dependent conductivity of single walled carbon nanotube(SWCNT) networks of various densities. The ac conductivity as a function of frequency is consistent with the extended pair approximation model and increases with frequency above an onset frequency $ω_0$ which varies over seven decades with a range of film thickness from sub-monolayer to 200 nm. The nonlinear electric field-dependent DC conductivity shows strong dependence on film thickness as well. Measurement of the electric field dependence of the resistance R(E) allows for the determination of a length scale $L_{E}$ possibly characterizing the distance between tube contacts, which is found to systematically decrease with increasing film thickness. The onset frequency $ω_0$ of ac conductivity and the length scale $L_{E}$ of SWCNT networks are found to be correlated, and a physically reasonable empirical formula relating them has been proposed. Such studies will help the understanding of transport properties and benefit the applications of this material system.

cond-mat.mtrl-sci↗