arXiv subjects

Dong Wang

Publications and source records attributed to Dong Wang.

At least 703 records · Page 39Linked to original sources

arXiv2018

Deep factorization for speech signal

Various informative factors mixed in speech signals, leading to great difficulty when decoding any of the factors. An intuitive idea is to factorize each speech frame into individual informative factors, though it turns out to be highly difficult. Recently, we found that speaker traits, which were assumed to be long-term distributional properties, are actually short-time patterns, and can be learned by a carefully designed deep neural network (DNN). This discovery motivated a cascade deep factorization (CDF) framework that will be presented in this paper. The proposed framework infers speech factors in a sequential way, where factors previously inferred are used as conditional variables when inferring other factors. We will show that this approach can effectively factorize speech signals, and using these factors, the original speech spectrum can be recovered with a high accuracy. This factorization and reconstruction approach provides potential values for many speech processing tasks, e.g., speaker recognition and emotion recognition, as will be demonstrated in the paper.

eess.AS↗

arXiv2018

Agile Amulet: Real-Time Salient Object Detection with Contextual Attention

This paper proposes an Agile Aggregating Multi-Level feaTure framework (Agile Amulet) for salient object detection. The Agile Amulet builds on previous works to predict saliency maps using multi-level convolutional features. Compared to previous works, Agile Amulet employs some key innovations to improve training and testing speed while also increase prediction accuracy. More specifically, we first introduce a contextual attention module that can rapidly highlight most salient objects or regions with contextual pyramids. Thus, it effectively guides the learning of low-layer convolutional features and tells the backbone network where to look. The contextual attention module is a fully convolutional mechanism that simultaneously learns complementary features and predicts saliency scores at each pixel. In addition, we propose a novel method to aggregate multi-level deep convolutional features. As a result, we are able to use the integrated side-output features of pre-trained convolutional networks alone, which significantly reduces the model parameters leading to a model size of 67 MB, about half of Amulet. Compared to other deep learning based saliency methods, Agile Amulet is of much lighter-weight, runs faster (30 fps in real-time) and achieves higher performance on seven public benchmarks in terms of both quantitative and qualitative evaluation.

cs.CV↗

arXiv2018

A diffusion generated method for computing Dirichlet partitions

A Dirichlet $k$-partition of a closed $d$-dimensional surface is a collection of $k$ pairwise disjoint open subsets such that the sum of their first Laplace-Beltrami-Dirichlet eigenvalues is minimal. In this paper, we develop a simple and efficient diffusion generated method to compute Dirichlet $k$-partitions for $d$-dimensional flat tori and spheres. For the $2d$ flat torus, for most values of $k=3$-9,11,12,15,16, and 20, we obtain hexagonal honeycombs. For the $3d$ flat torus and $k=2,4,8,16$, we obtain the rhombic dodecahedral honeycomb, the Weaire-Phelan honeycomb, and Kelvin's tessellation by truncated octahedra. For the $4d$ flat torus, for $k=4$, we obtain a constant extension of the rhombic dodecahedral honeycomb along the fourth direction and for $k=8$, we obtain a 24-cell honeycomb. For the $2d$ sphere, we also compute Dirichlet partitions for $k=3$-7,9,10,12,14,20. Our computational results agree with previous studies when a comparison is available. As far as we are aware, these are the first published results for Dirichlet partitions of the $4d$ flat torus.

math.OC↗

arXiv2018

Human and Machine Speaker Recognition Based on Short Trivial Events

Trivial events are ubiquitous in human to human conversations, e.g., cough, laugh and sniff. Compared to regular speech, these trivial events are usually short and unclear, thus generally regarded as not speaker discriminative and so are largely ignored by present speaker recognition research. However, these trivial events are highly valuable in some particular circumstances such as forensic examination, as they are less subjected to intentional change, so can be used to discover the genuine speaker from disguised speech. In this paper, we collect a trivial event speech database that involves 75 speakers and 6 types of events, and report preliminary speaker recognition results on this database, by both human listeners and machines. Particularly, the deep feature learning technique recently proposed by our group is utilized to analyze and recognize the trivial events, which leads to acceptable equal error rates (EERs) despite the extremely short durations (0.2-0.5 seconds) of these events. Comparing different types of events, 'hmm' seems more speaker discriminative.

cs.SD↗

arXiv2018

The Age of Social Sensing

Online social media, such as Twitter and Instagram, democratized information broadcast, allowing anyone to share information about themselves and their surroundings at an unprecedented scale. The large volume of information thus posted on these media offer a new lens into the physical world through the eyes of the social network. The exploitation of this lens to inspect aspects of world state has recently been termed social sensing. The power of manipulating reality via the use (or intentional misuse) of social media opened concerns with issues ranging from radicalization by terror propaganda to potential manipulation of elections in mature democracies. Many important challenges and open research questions arise in this emerging field that aims to better understand how information can be extracted from the medium and what properties characterize the extracted information and the world it represents. Addressing the above challenges requires multi-disciplinary research at the intersection of computer science and social sciences that combines cyber-physical computing, sociology, sensor networks, social networks, cognition, data mining, estimation theory, data fusion, information theory, linguistics, machine learning, behavioral economics, and possibly others. This paper surveys important directions in social sensing, identifies current research challenges, and outlines avenues for future research.

cs.SI↗

arXiv2018

Watching charge separation in nanoantennas by ultrafast point-projection electron microscopy

Watching the motion of electrons on their natural nanometre length- and femtosecond time scales is a fundamental goal and an open challenge of contemporary ultrafast science. Optical techniques and electron microscopy currently mostly provide either ultrahigh temporal or spatial resolution, yet, microscopy techniques with combined space-time resolution need further development. Here we create an ultrafast electron source by plasmon nanofocusing on a sharp gold taper and implement this source in an ultrafast point-projection electron microscope. This source is used, in an optical pump - electron probe experiment, to study ultrafast photoemission from a nanometer-sized plasmonic antenna. We show that the real space motion of the photoemitted electrons and residual holes in the metal is probed with 20-nm spatial resolution and 25-fs time resolution. This is a step forward towards time-resolved microscopy of electronic motion in nanostructures.

physics.optics↗

arXiv2018

Distributed Estimation of Principal Eigenspaces

Principal component analysis (PCA) is fundamental to statistical machine learning. It extracts latent principal factors that contribute to the most variation of the data. When data are stored across multiple machines, however, communication cost can prohibit the computation of PCA in a central location and distributed algorithms for PCA are thus needed. This paper proposes and studies a distributed PCA algorithm: each node machine computes the top $K$ eigenvectors and transmits them to the central server; the central server then aggregates the information from all the node machines and conducts a PCA based on the aggregated information. We investigate the bias and variance for the resulting distributed estimator of the top $K$ eigenvectors. In particular, we show that for distributions with symmetric innovation, the empirical top eigenspaces are unbiased and hence the distributed PCA is "unbiased". We derive the rate of convergence for distributed PCA estimators, which depends explicitly on the effective rank of covariance, eigen-gap, and the number of machines. We show that when the number of machines is not unreasonably large, the distributed PCA performs as well as the whole sample PCA, even without full access of whole data. The theoretical results are verified by an extensive simulation study. We also extend our analysis to the heterogeneous case where the population covariance matrices are different across local machines but share similar top eigen-structures.

stat.CO↗

arXiv2018

Intrinsic Charge Transport in Stanene: Roles of Bucklings and Electron-Phonon Couplings

The intrinsic charge transport of stanene is investigated by using density function theory and density function perturbation theory coupled with Boltzmann transport equations from first principles. The accurate Wannier interpolations are applied to calculate the charge carrier scatterings with all branches of phonons with dispersion contribution. The intrinsic carrier mobilities are predicted to be 2~3$\times10^3$ cm$^2$/(V s) at 300 K, and we find that the intervalley scatterings from the out-of-plane and transverse acoustic phonon modes dominate the carrier relaxation. In contrast, the intrinsic carrier mobilities obtained by the conventional deformation potential approach (Long et al., J. Am. Chem. Soc. 2009, 131, 17728) are found to as large as 2~3$\times$10$^6$ cm$^2$/(V s) at 300 K, in which the longitudinal acoustic phonons are assumed to be the only scattering mechanism. The inadequacy of the deformation potential approximation in stanene is attributed to the buckling of the honeycomb structure, which originates from the $sp^2-sp^3$ orbital hybridization and results in broken mirror symmetry as compared to graphene. The high carrier mobility of stanene renders it a promising candidate in nanoelectronics and spintronics applications and we propose to enhance its carrier mobilities by suppressing the out-of-plane vibrations by substrate suspension or clamping.

physics.chem-ph↗

arXiv2017

Structure Re-determination and Superconductivity Observation of Bulk 1T MoS2

2H MoS2 has been intensively studied because of layer-dependent electronic structures and novel physical properties. Though the metastable 1T MoS2 with the [MoS6] octahedron was observed from the microscopic area, the true crystal structure of 1T phase has not been determined strictly. Moreover, the true physical properties have not been demonstrated from experiments due to the challenge for the preparation of pure 1T MoS2 crystals. Here, we successfully synthesized the 1T MoS2 single crystals and re-determined the crystal structure of 1T MoS2 from single-crystal X-ray diffraction. 1T MoS2 crystalizes in space group P-3m1 with a cell of a = b = 3.190(3) Å and c = 5.945(6) Å. The individual MoS2 layer consists of MoS6 octahedron sharing edge with each other. More surprisingly, the bulk 1T MoS2 crystals undergo a superconducting transition of Tc = 4 K, which is the first observation of superconductivity in pure 1T MoS2 phase.

cond-mat.supr-con↗

arXiv2017

Software Engineering Solutions To Support Vertical Transportation

In this paper we introduce the core results of the project on visualisation and analysis of data collected from the vertical transport facilities. The aim of the project was to provide better user experience as well as to help building maintenance staff to increase productivity of their work. We elaborated a web-based system for vertical transportation, to cover the needs of (1) staff working on building maintenance, (2) people who are regularly using the facilities in the corresponding buildings.

cs.HC↗

arXiv2017

`Sinking' in a bed of grains activated by shearing

We show how a weak force, $f$, enables intruder motion through dense granular materials subject to external mechanical excitations, in the present case stepwise shearing. A force acts on a Teflon disc in a two dimensional system of photoelastic discs. This force is much smaller than the smallest force needed to move the disc without any external excitation. In a cycle, material + intruder are sheared quasi-statically from $γ= 0$ to $γ_{max}$, and then backwards to $γ= 0$. During various cycle phases, fragile and jammed states form. Net intruder motion, $δ$, occurs during fragile periods generated by shear reversals. $δ$ per cycle, e.g. the quasistatic rate $c$, is constant, linearly dependent on $γ_{max}$ and $f$. It vanishes as, $c \propto (ϕ_c - ϕ)^a$, with $a \simeq 3$ and $ϕ_c \simeq ϕ_J$, reflecting the stiffening of granular systems under shear as $ϕ\rightarrow ϕ_J$. The intruder motion induces large scale grain circulation. In the intruder frame, this motion is a granular analogue to fluid flow past a cylinder, where $f$ is the drag force exerted by the flow.

cond-mat.soft↗

arXiv2017

Medical Diagnosis From Laboratory Tests by Combining Generative and Discriminative Learning

A primary goal of computational phenotype research is to conduct medical diagnosis. In hospital, physicians rely on massive clinical data to make diagnosis decisions, among which laboratory tests are one of the most important resources. However, the longitudinal and incomplete nature of laboratory test data casts a significant challenge on its interpretation and usage, which may result in harmful decisions by both human physicians and automatic diagnosis systems. In this work, we take advantage of deep generative models to deal with the complex laboratory tests. Specifically, we propose an end-to-end architecture that involves a deep generative variational recurrent neural networks (VRNN) to learn robust and generalizable features, and a discriminative neural network (NN) model to learn diagnosis decision making, and the two models are trained jointly. Our experiments are conducted on a dataset involving 46,252 patients, and the 50 most frequent tests are used to predict the 50 most common diagnoses. The results show that our model, VRNN+NN, significantly (p<0.001) outperforms other baseline models. Moreover, we demonstrate that the representations learned by the joint training are more informative than those learned by pure generative models. Finally, we find that our model offers a surprisingly good imputation for missing values.

cs.AI↗

arXiv2017

A generalized MBO diffusion generated motion for orthogonal matrix-valued fields

We consider the problem of finding stationary points of the Dirichlet energy for orthogonal matrix-valued fields. Following the Ginzburg-Landau approach, this energy is relaxed by penalizing the matrix-valued field when it does not take orthogonal matrix values. A generalization of the MBO diffusion generated motion is introduced that effectively finds local minimizers of this energy by iterating two steps until convergence. In the first step, as in the original method, the current matrix-valued field is evolved by the diffusion equation. In the second step, the field is pointwise reassigned to the closest orthogonal matrix, which can be computed via the singular value decomposition. We extend the Lyapunov function of Esedoglu and Otto to show that the method is non-increasing on iterates and hence, unconditionally stable. We also prove that spatially discretized iterates converge to a stationary solution in a finite number of iterations. The algorithm is implemented using the closest point method and non-uniform fast Fourier transform. We conclude with several numerical experiments on flat tori and closed surfaces, which, unsurprisingly, exhibit classical behavior from the Allen-Cahn and complex Ginzburg Landau equations, but also new phenomena.

math.AP↗

arXiv2017

Modeling synthetic spectra for transiting extrasolar giant planets: detectability of H$_2$S and PH$_3$ with JWST

JWST$'$s large aperture and wide wavelength coverage will enable it to collect the highest quality transit spectra observed so far. For exoplanetary atmospheres we expect to retrieve the abundance of the most abundant molecules, such as H$_2$O, CO, and CH$_4$. Other molecules, such as H$_2$S and PH$_3$, have been observed in Jupiter and Saturn but their chemistry and detectability in strongly irradiated planets is highly unknown. In this paper, we make the first effort to study their spectral features in solar composition atmospheres, and evaluate their detectability with JWST. We model the chemistry of phosphorus and sulfur in solar composition atmospheres. Our model includes the effect of vertical transport. Photochemistry effects are not included in our calculations. Using the abundance profiles, we model the JWST transmission and emission spectra for K=6.8 G-type star and for planets with cloud-free solar composition atmospheres. We find PH$_3$ is detectable at 3 sigma from transmission spectra of the simulated atmosphere with $T_{\rm eq}$ $<$ 500K using the NIRCam LW grism F444W mode with a total observing time of 28.8 hrs. H$_2$S is detectable at 3 sigma in the transmission and emission spectra for the simulated planet with $T_{\rm eq}$ $>$ 1500K using the NIRCam LW grism F322W2 mode with a total observing time of 24.0 hrs. Our results specifically highlight the importance of including H$_2$S for future abundances retrieval with JWST. The presence of clouds and hazes challenges the detections of PH$_3$ and H$_2$S, but H$_2$S features are still expected to be present in the emission spectra.

astro-ph.EP↗

arXiv2017

Enhanced Neural Machine Translation by Learning from Draft

Neural machine translation (NMT) has recently achieved impressive results. A potential problem of the existing NMT algorithm, however, is that the decoding is conducted from left to right, without considering the right context. This paper proposes an two-stage approach to solve the problem. In the first stage, a conventional attention-based NMT system is used to produce a draft translation, and in the second stage, a novel double-attention NMT system is used to refine the translation, by looking at the original input as well as the draft translation. This drafting-and-refinement can obtain the right-context information from the draft, hence producing more consistent translations. We evaluated this approach using two Chinese-English translation tasks, one with 44k pairs and 1M pairs respectively. The experiments showed that our approach achieved positive improvements over the conventional NMT system: the improvements are 2.4 and 0.9 BLEU points on the small-scale and large-scale tasks, respectively.

cs.CL↗

arXiv2017

Recovering the lost steerability of quantum states within non-Markovian environments by utilizing quantum partially collapsing measurements

In this Letter, we mainly investigate the dynamic behavior of quantum steering and how to effectively recover the lost steerability of quantum states within non-Markovian environments. We consider two different cases (one-subsystem or all-subsystem interacts with the dissipative environments), and obtain that the dynamical interaction between system initialized by a Werner state and the non-Markovian environments can induce the quasi-periodic quantum entanglement (concurrence) resurgence, however, quantum steering cannot retrieve in such a condition. And we can obtain that the resurgent quantum entanglement cannot be utilized to achieve quantum steering. Subsequently, we put forward a feasible physical scheme for recovering the steerability of quantum states within the non-Markovian noises by prior weak measurement on each subsystem before the interaction with dissipative environments followed by post weak measurement reversal. It is shown that the steerability of quantum states and the fidelity can be effectively restored. Furthermore, the results show that the larger the weak measurement strength is, the better the effectiveness of the scheme is. Consequently, our investigations might be beneficial to recover the lost steerability of quantum states within the non-Markovian regimes.

quant-ph↗

arXiv2017

Design, Modeling and Dynamic Compensation PID Control of a Fully-Actuated Aerial Manipulation System

This paper addresses design, modeling and dynamic-compensation PID (dc-PID) control of a novel type of fully-actuated aerial manipulation (AM) system. Firstly, design of novel mechanical structure of the AM is presented. Secondly, kinematics and dynamics of AM are modeled using Craig parameters and recursion Newton-Euler equations respectively, which give rise to a more accurate dynamic relationship between aerial platform and manipulator. Then, the dynamic-compensation PID control is proposed to solve the problem of fully-actuated control of AM. Finally, uniform coupled matrix equations between driving forces/moments and rotor speeds are derived, which can support design and analysis of parameters and decoupling theoretically. It is taken into account practical problems including noise and perturbation, parameter uncertainty, and power limitation in simulations, and results from simulations shows that the AM system presented can be fully-actued controlled with advanced control performances, which can not achieved theoretically in traditional AM. And with compared to backstepping control dc-PID has better control accuracy and capability to disturbance rejection in two simulations of aerial operation tasks with motion of joint. The experiment of dc-pid proves the availability and effectiveness of the method proposed.

cs.RO↗

arXiv2017

Local and global avalanches in a 2D sheared granular medium

We present the experimental and numerical studies of a 2D sheared amorphous material constituted of bidisperse photo-elastic disks. We analyze the statistics of avalanches during shear including the local and global fluctuations in energy and changes in particle positions and orientations. We find scale free distributions for these global and local avalanches denoted by power-laws whose cut-offs vary with inter-particle friction and packing fraction. Different exponents are found for these power-laws depending on the quantity from which variations are extracted. An asymmetry in time of the avalanche shapes is evidenced along with the fact that avalanches are mainly triggered from the shear bands. A simple relation independent from the intensity, is found between the number of local avalanches and the global avalanches they form. We also compare these experimental and numerical results for both local and global fluctuations to predictions from meanfield and depinning theories.

cond-mat.soft↗