Searcharxiv⌕ Search

arXiv subjects

Sen Yang

Publications and source records attributed to Sen Yang.

At least 199 records · Page 11Linked to original sources
arXiv2020

What Have We Achieved on Text Summarization?

Deep learning has led to significant improvement in text summarization with various methods investigated and improved ROUGE scores reported over the years. However, gaps still exist between summaries produced by automatic summarizers and human professionals. Aiming to gain more understanding of summarization systems with respect to their strengths and limits on a fine-grained syntactic and semantic level, we consult the Multidimensional Quality Metric(MQM) and quantify 8 major sources of errors on 10 representative summarization models manually. Primarily, we find that 1) under similar settings, extractive summarizers are in general better than their abstractive counterparts thanks to strength in faithfulness and factual-consistency; 2) milestone techniques such as copy, coverage and hybrid extractive/abstractive methods do bring specific improvements but also demonstrate limitations; 3) pre-training techniques, and in particular sequence-to-sequence pre-training, are highly effective for improving text summarization, with BART giving the best results.

cs.CL↗
arXiv2020

Spherical Accretion Flow onto General Parameterized Spherically Symmetric Black Hole Spacetimes

The transonic phenomenon of black hole accretion and the existence of the photon sphere are the characteristics of strong gravitational fields near a black hole horizon. In this work, we study spherical accretion flow onto a general parametrized spherically symmetric black hole spacetimes. For this purpose, we analyze the accretion process of various perfect fluids, such as the isothermal fluid of ultra-stiff, ultra-relativistic, and sub-relativistic types and polytropic fluid, respectively. The influences of extra parameters beyond the Schwarzschild black hole in the general parameterized spherically symmetric black hole on the flow behaviors of the above-mentioned test fluids are studied in detail. In addition, by studying the accretion of ideal photon gas, we further discuss the correspondence between the sonic radius of accreting photon gas and the photon sphere for the general parameterized spherically symmetric black hole. Some possible future extensions of our analysis are also discussed.

gr-qc↗
arXiv2020

Leveraging Adversarial Training in Self-Learning for Cross-Lingual Text Classification

In cross-lingual text classification, one seeks to exploit labeled data from one language to train a text classification model that can then be applied to a completely different language. Recent multilingual representation models have made it much easier to achieve this. Still, there may still be subtle differences between languages that are neglected when doing so. To address this, we present a semi-supervised adversarial training process that minimizes the maximal loss for label-preserving input perturbations. The resulting model then serves as a teacher to induce labels for unlabeled target language samples that can be used during further adversarial training, allowing us to gradually adapt our model to the target language. Compared with a number of strong baselines, we observe significant gains in effectiveness on document and intent classification for a diverse set of languages.

cs.CL↗
arXiv2020

Event Arguments Extraction via Dilate Gated Convolutional Neural Network with Enhanced Local Features

Event Extraction plays an important role in information-extraction to understand the world. Event extraction could be split into two subtasks: one is event trigger extraction, the other is event arguments extraction. However, the F-Score of event arguments extraction is much lower than that of event trigger extraction, i.e. in the most recent work, event trigger extraction achieves 80.7%, while event arguments extraction achieves only 58%. In pipelined structures, the difficulty of event arguments extraction lies in its lack of classification feature, and the much higher computation consumption. In this work, we proposed a novel Event Extraction approach based on multi-layer Dilate Gated Convolutional Neural Network (EE-DGCNN) which has fewer parameters. In addition, enhanced local information is incorporated into word features, to assign event arguments roles for triggers predicted by the first subtask. The numerical experiments demonstrated significant performance improvement beyond state-of-art event extraction approaches on real-world datasets. Further analysis of extraction procedure is presented, as well as experiments are conducted to analyze impact factors related to the performance improvement.

cs.CL↗
arXiv2020

Diverse electronic and magnetic properties of CrS2 enabling novel strain-controlled 2D lateral heterostructure spintronic devices

Lateral heterostructures of two-dimensional (2D) materials, integrating different phases or materials into a single piece of nanosheet, have attracted intensive research interests in the past few years for high-performance electronic and optoelectronic devices. It also holds promises to significantly improve the performance and enable new functions of spintronic devices. It is imperative to have a 2D material possessing diverse electronic and magnetic properties that are required in spintronics. In this work, using density functional theory calculations, we surveyed all IV, V and VI group transition metal dichalcogenides (TMDs) and discovered that CrS2 has the most diverse electronic and magnetic properties: antiferromagnetic (AFM) metallic 1T phase, nonmagnetic (NM) semiconductor 2H phase, and ferromagnetic (FM) semiconductor 1T_prime phase with a Curie temperature of ~1000 K. More interestingly, we found that a tensile or compressive strain could turn 1T_prime phase into a spin-up or spin-down half metal. Such a unique feature enables designing strain-controlled spintronic devices using a single piece of CrS2 crystal with improved energy efficiency, which remains a challenge in miniaturization of spintronic devices. In-depth analysis attributed the unique strain tunability to the interplay between strain-induced lattice deformation and different spatial orientation of the spin-up/spin-down electronic orbitals. A prototypical design of a simple spin-valve logic device operated by strain is also presented.

cond-mat.mtrl-sci↗
arXiv2020

Probing local pressure environment in anvil cells with nitrogen vacancy (NV-) centers in diamond

Important discoveries have frequently been made through the studies of matter under high pressure. The conditions of the pressure environment are important for the interpretation of the experimental results. Due to various restrictions inside the pressure cell, detailed information relevant to the pressure environment, such as the pressure distribution, can be hard to obtain experimentally. Here we present the study of pressure distributions inside the pressure medium under different experimental conditions with NV centers in diamond particles as the sensor. These studies not only show a good spatial resolution, wide temperature and pressure working ranges, compatibility of the existing pressure cell design with the new method, but also demonstrate the usefulness to measure with these sensors as the pressure distribution is sensitive to various factors. The method and the results will benefit many disciplines such as material research and phase transitions in fluid dynamics.

cond-mat.supr-con↗
arXiv2020

SK-Unet: an Improved U-net Model with Selective Kernel for the Segmentation of Multi-sequence Cardiac MR

In the clinical environment, myocardial infarction (MI) as one com-mon cardiovascular disease is mainly evaluated based on the late gadolinium enhancement (LGE) cardiac magnetic resonance images (CMRIs). The auto-matic segmentations of left ventricle (LV), right ventricle (RV), and left ven-tricular myocardium (LVM) in the LGE CMRIs are desired for the aided diag-nosis in clinic. To accomplish this segmentation task, this paper proposes a modified U-net architecture by combining multi-sequence CMRIs, including the cine, LGE, and T2-weighted CMRIs. The cine and T2-weighted CMRIs are used to assist the segmentation in the LGE CMRIs. In this segmentation net-work, the squeeze-and-excitation residual (SE-Res) and selective kernel (SK) modules are inserted in the down-sampling and up-sampling stages, respective-ly. The SK module makes the obtained feature maps more informative in both spatial and channel-wise space, and attains more precise segmentation result. The utilized dataset is from the MICCAI challenge (MS-CMRSeg 2019), which is acquired from 45 patients including three CMR sequences. The cine and T2-weighted CMRIs acquired from 35 patients and the LGE CMRIs acquired from 5 patients are labeled. Our method achieves the mean dice score of 0.922 (LV), 0.827 (LVM), and 0.874 (RV) in the LGE CMRIs.

eess.IV↗
arXiv2019

Modulating Peierls distortion of 1T_prime MoS2 via charge doping: a new charge density wave phase, reversible phase transition, and excellent electromechanical properties

The 1T_prime phase of transition metal dichalcogenides (TMDs) is a low symmetry charge density wave (CDW) phase, which can be regarded as a periodically distorted structure (Peierls distortion) of the high symmetry 1T phase. In this paper, using density functional theory (DFT) calculations, we report that the positive charge (hole) injection is an effective method to modulate the Peierls distortion of MoS2 1T_prime for new CDW phase and superior electromechanical properties. A new stable CDW phase is discovered at hole doping level of ~ 0.10h+/atom, named as 1Tt_prime. The hole charging and discharging can induce a reversible phase transition of MoS2 among the three phases, 1T, 1T_prime and 1Tt_prime. Such reversible phase transition leads to superior electromechanical properties including a strain output as high as -5.8% with a small hysteresis loop, multi-step super-elasticity, and multi-shape memory effect, which are valuable in active electromechanical device designs at nanoscale. In-depth analysis on the change of electronic structure under hole doping was performed to understand the new CDW phase and the observed phase transition. Using charge doping to modulate the Peierls distortion in two-dimensional materials can serve as a general concept for nano-active material designs.

cond-mat.mtrl-sci↗
arXiv2019

Fabrication of BaZrS3 chalcogenide perovskite thin films for optoelectronics

BaZrS3 is a prototypical chalcogenide perovskite, an emerging class of unconventional semiconductor. Recent results on powder samples reveal that it is a material with a direct band gap of 1.7-1.8 eV, a very strong light-matter interaction, and a high chemical stability. However, many of the fundamental properties are unknown, hindering the ability to apply BaZrS3 for optoelectronics. Here we report the fabrication of BaZrS3 thin films, by sulfurization of oxide films deposited by pulsed laser deposition. We show that these films are n-type with carrier densities in the range of 10^19-10^20 cm^-3. Depending on the processing temperature, the Hall mobility ranges from 2.1 to 13.7 cm^2/Vs. The absorption coefficient is > 10^5 cm-1 at photon energy > 1.97 eV. Temperature dependent conductivity measurements suggest shallow donor levels. These results assure that BaZrS3 is a promising candidate for optoelectronics such as photodetectors, photovoltaics, and light emitting diodes.

cond-mat.mtrl-sci↗
arXiv2019

Machine learning magnetic parameters from spin configurations

Hamiltonian parameter estimation is crucial in condensed matter physics, but time and cost consuming in terms of resources used. With advances in observation techniques, high-resolution images with more detailed information are obtained, which can serve as an input to machine learning (ML) algorithms to extract Hamiltonian parameters. However, the number of labeled images is rather limited. Here, we provide a protocol for Hamiltonian parameter estimation based on a machine learning architecture, which is trained on a small amount of simulated images and applied to experimental spin configuration images. Sliding windows on the input images enlarges the number of training images; therefore we can train well a neural network on a small dataset of simulated images which are generated adaptively using the same external conditions such as temperature and magnetic field as the experiment. The neural network is applied to the experimental image and estimates magnetic parameters efficiently. We demonstrate the success of the estimation by reproducing the same configuration from simulation and predict a hysteresis loop accurately. Our approach paves a way to a stable and general parameter estimation.

cond-mat.dis-nn↗
arXiv2019

On the image of the Abel-Jacobi map

We prove and generalize an observation of Green and Griffiths on the infinitesimal form of the Abel-Jacobi map. As an application, we prove that the infinitesimal form of a conjecture by Griffiths and Harris is true.

math.AG↗
arXiv2019

On the Linear Speedup Analysis of Communication Efficient Momentum SGD for Distributed Non-Convex Optimization

Recent developments on large-scale distributed machine learning applications, e.g., deep neural networks, benefit enormously from the advances in distributed non-convex optimization techniques, e.g., distributed Stochastic Gradient Descent (SGD). A series of recent works study the linear speedup property of distributed SGD variants with reduced communication. The linear speedup property enable us to scale out the computing capability by adding more computing nodes into our system. The reduced communication complexity is desirable since communication overhead is often the performance bottleneck in distributed systems. Recently, momentum methods are more and more widely adopted in training machine learning models and can often converge faster and generalize better. For example, many practitioners use distributed SGD with momentum to train deep neural networks with big data. However, it remains unclear whether any distributed momentum SGD possesses the same linear speedup property as distributed SGD and has reduced communication complexity. This paper fills the gap by considering a distributed communication efficient momentum SGD method and proving its linear speedup property.

math.OC↗
arXiv2019

Shrinking the Upper Confidence Bound: A Dynamic Product Selection Problem for Urban Warehouses

The recent rising popularity of ultra-fast delivery services on retail platforms fuels the increasing use of urban warehouses, whose proximity to customers makes fast deliveries viable. The space limit in urban warehouses poses a problem for the online retailers: the number of products (SKUs) they carry is no longer "the more, the better", yet it can still be significantly large, reaching hundreds or thousands in a product category. In this paper, we study algorithms for dynamically identifying a large number of products (i.e., SKUs) with top customer purchase probabilities on the fly, from an ocean of potential products to offer on retailers' ultra-fast delivery platforms. We distill the product selection problem into a semi-bandit model with linear generalization. There are in total $N$ different arms, each with a feature vector of dimension $d$. The player pulls $K$ arms in each period and observes the bandit feedback from each of the pulled arms. We focus on the setting where $K$ is much greater than the number of total time periods $T$ or the dimension of product features $d$. We first analyze a standard UCB algorithm and show its regret bound can be expressed as the sum of a $T$-independent part $\tilde O(K d^{3/2})$ and a $T$-dependent part $\tilde O(d\sqrt{KT})$, which we refer to as "fixed cost" and "variable cost" respectively. To reduce the fixed cost for large $K$ values, we propose a novel online learning algorithm, which iteratively shrinks the upper confidence bounds within each period, and show its fixed cost is reduced by a factor of $d$ to $\tilde O(K \sqrt{d})$. Moreover, we test the algorithms on an industrial dataset from Alibaba Group. Experimental results show that our new algorithm reduces the total regret of the standard UCB algorithm by at least 10%.

math.OC↗
arXiv2019

SERENADE: A Parallel Randomized Algorithm Suite for Crossbar Scheduling in Input-Queued Switches

Most of today's high-speed switches and routers adopt an input-queued crossbar switch architecture. Such a switch needs to compute a matching (crossbar schedule) between the input ports and output ports during each switching cycle (time slot). A key research challenge in designing large (in number of input/output ports $N$) input-queued crossbar switches is to develop crossbar scheduling algorithms that can compute "high quality" matchings -- i.e., those that result in high switch throughput (ideally $100\%$) and low queueing delays for packets -- at line rates. SERENA is one such algorithm: it outputs excellent matching decisions that result in $100\%$ switch throughput and reasonably good queueing delays. However, since SERENA is a centralized algorithm with $O(N)$ computational complexity, it cannot support switches that both are large and have a very high line rate per port. In this work, we propose SERENADE (SERENA, the Distributed Edition), a parallel iterative algorithm that emulates SERENA in only $O(\log N)$ iterations between input ports and output ports, and hence has a time complexity of only $O(\log N)$ per port. We prove that SERENADE can exactly emulate SERENA. We also propose an early-stop version of SERENADE, called O-SERENADE, to only approximately emulate SERENA. Through extensive simulations, we show that O-SERENADE can achieve 100\% throughput and that it has similar as or slightly better delay performance than SERENA under various load conditions and traffic patterns.

cs.PF↗
arXiv2019

On the Convergence of (Stochastic) Gradient Descent with Extrapolation for Non-Convex Optimization

Extrapolation is a well-known technique for solving convex optimization and variational inequalities and recently attracts some attention for non-convex optimization. Several recent works have empirically shown its success in some machine learning tasks. However, it has not been analyzed for non-convex minimization and there still remains a gap between the theory and the practice. In this paper, we analyze gradient descent and stochastic gradient descent with extrapolation for finding an approximate first-order stationary point in smooth non-convex optimization problems. Our convergence upper bounds show that the algorithms with extrapolation can be accelerated than without extrapolation.

math.OC↗
arXiv2018

Quantum sensing of local magnetic field texture in strongly correlated electron systems under extreme conditions

An important feature of strong correlated electron systems is the tunability between interesting ground states such as unconventional superconductivity and exotic magnetism. Pressure is a clean, continuous and systematic tuning parameter. However, due to the restricted accessibility introduced by high-pressure devices, compatible magnetic field sensors with sufficient sensitivity are rare. This greatly limits the detections and detailed studies of pressure-induced phenomena. Here, we utilize nitrogen vacancy (NV) centers in diamond as a powerful, spatially-resolved vector field sensor for material research under pressure at cryogenic temperatures. Using a single crystal of BaFe2(As0:59P0:41)2 as an example, we extract the superconducting transition temperature (Tc), the local magnetic field profile in the Meissner state and the critical fields (Hc1 and Hc2). The method developed in this work will become a unique tool for tuning, probing and understanding quantum many body systems.

cond-mat.mes-hall↗
arXiv2018

Parallel Restarted SGD with Faster Convergence and Less Communication: Demystifying Why Model Averaging Works for Deep Learning

In distributed training of deep neural networks, parallel mini-batch SGD is widely used to speed up the training process by using multiple workers. It uses multiple workers to sample local stochastic gradient in parallel, aggregates all gradients in a single server to obtain the average, and update each worker's local model using a SGD update with the averaged gradient. Ideally, parallel mini-batch SGD can achieve a linear speed-up of the training time (with respect to the number of workers) compared with SGD over a single worker. However, such linear scalability in practice is significantly limited by the growing demand for gradient communication as more workers are involved. Model averaging, which periodically averages individual models trained over parallel workers, is another common practice used for distributed training of deep neural networks since (Zinkevich et al. 2010) (McDonald, Hall, and Mann 2010). Compared with parallel mini-batch SGD, the communication overhead of model averaging is significantly reduced. Impressively, tremendous experimental works have verified that model averaging can still achieve a good speed-up of the training time as long as the averaging interval is carefully controlled. However, it remains a mystery in theory why such a simple heuristic works so well. This paper provides a thorough and rigorous theoretical study on why model averaging can work as well as parallel mini-batch SGD with significantly less communication overhead.

math.OC↗