arXiv subjects

Tao Wang

Publications and source records attributed to Tao Wang.

At least 721 records · Page 40Linked to original sources

arXiv2020

Making Robots Draw A Vivid Portrait In Two Minutes

Significant progress has been made with artistic robots. However, existing robots fail to produce high-quality portraits in a short time. In this work, we present a drawing robot, which can automatically transfer a facial picture to a vivid portrait, and then draw it on paper within two minutes averagely. At the heart of our system is a novel portrait synthesis algorithm based on deep learning. Innovatively, we employ a self-consistency loss, which makes the algorithm capable of generating continuous and smooth brush-strokes. Besides, we propose a componential sparsity constraint to reduce the number of brush-strokes over insignificant areas. We also implement a local sketch synthesis algorithm, and several pre- and post-processing techniques to deal with the background and details. The portrait produced by our algorithm successfully captures individual characteristics by using a sparse set of continuous brush-strokes. Finally, the portrait is converted to a sequence of trajectories and reproduced by a 3-degree-of-freedom robotic arm. The whole portrait drawing robotic system is named AiSketcher. Extensive experiments show that AiSketcher can produce considerably high-quality sketches for a wide range of pictures, including faces in-the-wild and universal images of arbitrary content. To our best knowledge, AiSketcher is the first portrait drawing robot that uses neural style transfer techniques. AiSketcher has attended a quite number of exhibitions and shown remarkable performance under diverse circumstances.

cs.RO↗

arXiv2020

Deep Learning based OTDOA Positioning for NB-IoT Communication Systems

Positioning is becoming a key component in many Internet of Things (IoT) applications. The main challenges and limitations are the narrow bandwidth, low power and low cost which reduces the accuracy of the time of arrival (TOA) estimation. In this paper, we consider the positioning scenario of Narrowband IoT (NB-IoT) that can benefit from observed time difference of arrival (OTDOA). By applying the deep learning based technique, we explore the generalization and feature extraction abilities of neural networks to tackle the aforementioned challenges. As demonstrated in the numerical experiments, the proposed algorithm can be used in different inter-site distance situations and results in a 15% and 50% positioning accuracy improvement compared with Gauss-Newton method in line-of-sight (LOS) scenario and non-line-of-sight (NLOS) scenario respectively.

eess.SP↗

arXiv2020

ALMA Deep Field in SSA22: A near-infrared-dark submillimeter galaxy at z=4.0

Deep surveys with Atacama Large Millimeter Array (ALMA) have uncovered a population of dusty star-forming galaxies which are faint or even undetected at optical to near-infrared wavelengths. Their faintness at short wavelengths makes detailed characterization of the population challenging. Here we present a spectroscopic redshift identification and characterization of one of such near-infrared-dark galaxy discovered by an ALMA deep survey. Detection of [CI](1-0) and CO(4-3) emission lines determines the precise redshift of the galaxy, ADF22.A2, to be z=3.9913+/-0.0008. On the basis of multi-wavelength analysis, ADF22.A2 is found to be a massive, star-forming galaxy with stellar mass Mstar = $1.1_{-0.6}^{+1.3}$ x 10^{11} Msun and SFR = $430_{-150}^{+230}$ Msun/yr. The molecular gas mass is derived to be M ($H_2$) = 5.9 +/- 1.5x10^{10} Msun, indicating a gas fraction of ~35%, and the ratios of $L_{\rm [CI](1-0)}/L_{\rm IR}$ and $L_{\rm [CI](1-0)}/L_{\rm CO(4-3)}$ suggests that the nature of the interstellar medium in ADF22.A2 is in accordance with those of other bright submillimeter galaxies. The properties of ADF22.A2, including redshift, star-formation rate, stellar mass, and depletion time scale (tau~0.1-0.2 Gyr), also suggest that ADF22.A2 has the characteristics expected for the progenitors of quiescent galaxies at z>3. Our results demonstrate the power of ALMA contiguous mapping and line scan to obtain an unbiased view of galaxy formation in the early Universe.

astro-ph.GA↗

arXiv2020

A Multi-Level Approach to Waste Object Segmentation

We address the problem of localizing waste objects from a color image and an optional depth image, which is a key perception component for robotic interaction with such objects. Specifically, our method integrates the intensity and depth information at multiple levels of spatial granularity. Firstly, a scene-level deep network produces an initial coarse segmentation, based on which we select a few potential object regions to zoom in and perform fine segmentation. The results of the above steps are further integrated into a densely connected conditional random field that learns to respect the appearance, depth, and spatial affinities with pixel-level accuracy. In addition, we create a new RGBD waste object segmentation dataset, MJU-Waste, that is made public to facilitate future research in this area. The efficacy of our method is validated on both MJU-Waste and the Trash Annotation in Context (TACO) dataset.

cs.CV↗

arXiv2020

Dynamics of reversed shear Alfvén eigenmode and energetic particles during current ramp-up

Hybrid MHD-gyrokinetic code simulations are used to investigate the dynamics of frequency sweeping reversed shear Alfvén eigenmode (RSAE) strongly driven by energetic particles (EPs) during plasma current ramp-up in a conventional tokamak configuration. A series of weakly reversed shear equilibria representing time slices of long timescale MHD equilibrium evolution is considered, where the self-consistent RSAE-EP resonant interactions on the short timescale are analyzed in detail. Both linear and nonlinear RSAE dynamics are shown to be subject to the non-perturbative effect of EPs by maximizing wave-EP power transfer. In linear stage, EPs induce evident mode structure and frequency shifts; meanwhile, RSAE saturates by radial decoupling with resonant EPs due to weak magnetic shear, and gives rise to global EP convective transport and non-adiabatic frequency chirping. The spatiotemporal scales of phase space wave-EP interactions are characterized by the perpendicular wavelength and wave-particle trapping time. The simulations provide insights into general as well as specific features of RSAE spectra and EP transport from experimental observations, and illustrate the fundamental physics of wave-EP resonant interaction with the interplay of magnetic geometry, plasma non-uniformity and non-perturbative EPs.

physics.plasm-ph↗

arXiv2020

Overcoming Classifier Imbalance for Long-tail Object Detection with Balanced Group Softmax

Solving long-tail large vocabulary object detection with deep learning based models is a challenging and demanding task, which is however under-explored.In this work, we provide the first systematic analysis on the underperformance of state-of-the-art models in front of long-tail distribution. We find existing detection methods are unable to model few-shot classes when the dataset is extremely skewed, which can result in classifier imbalance in terms of parameter magnitude. Directly adapting long-tail classification models to detection frameworks can not solve this problem due to the intrinsic difference between detection and classification.In this work, we propose a novel balanced group softmax (BAGS) module for balancing the classifiers within the detection frameworks through group-wise training. It implicitly modulates the training process for the head and tail classes and ensures they are both sufficiently trained, without requiring any extra sampling for the instances from the tail classes.Extensive experiments on the very recent long-tail large vocabulary object recognition benchmark LVIS show that our proposed BAGS significantly improves the performance of detectors with various backbones and frameworks on both object detection and instance segmentation. It beats all state-of-the-art methods transferred from long-tail image classification and establishes new state-of-the-art.Code is available at https://github.com/FishYuLi/BalancedGroupSoftmax.

cs.CV↗

arXiv2020

Gravity Probe Spin: Prospects for measuring general-relativistic precession of intrinsic spin using a ferromagnetic gyroscope

An experimental test at the intersection of quantum physics and general relativity is proposed: measurement of relativistic frame dragging and geodetic precession using intrinsic spin of electrons. The behavior of intrinsic spin in spacetime dragged and warped by a massive rotating body is an experimentally open question, hence the results of such a measurement could have important theoretical consequences. Such a measurement is possible by using mm-scale ferromagnetic gyroscopes in orbit around the Earth. Under conditions where the rotational angular momentum of a ferromagnet is sufficiently small, a ferromagnet's angular momentum is dominated by atomic electron spins and is predicted to exhibit macroscopic gyroscopic behavior. If such a ferromagnetic gyroscope is sufficiently isolated from the environment, rapid averaging of quantum uncertainty via the spin-lattice interaction enables readout of the ferromagnetic gyroscope dynamics with sufficient sensitivity to measure both the Lense-Thirring (frame dragging) and de Sitter (geodetic precession) effects due to the Earth.

gr-qc↗

arXiv2020

Double Circulant Self-dual Codes on Sextic Cyclotomy

This paper contributes to construct double circulant self-dual codes by sextic cyclotomy. Generator matrixes of a family of pure double circulant codes and a family of double circulant codes with boundary are formed from sextic cyclotomic classes. These codes are proved to be self-dual on certain conditions. Moreover, experiments show that some of them are optimal self-dual codes or best than ever codes over GF}(2) and GF}(4).

cs.IT↗

arXiv2020

A system to test 2D optoelectronic devices in high vacuum

The exploration of electronic and optoelectronic properties of two-dimensional (2D) materials has become one of the most attractive line of research since the isolation of graphene. Such 'all-surface materials' present a strong sensitivity to environmental conditions and thus characterization of the devices based on these materials usually requires measurement systems operating in high-vacuum. However, conventional optoelectronic probe-station testing systems are are not compatible with high vacuum operation and vacuum-compatible versions are rather expensive. Here, we present a high-vacuum system specifically designed to test electronic and optoelectronic devices based on 2D materials. This system can be implemented with low budget and it is mostly based on the assembly of commercially available standard vacuum and optic components. Despite the simplicity of this system we demonstrate full capabilities to characterize optoelectronic devices in a broad range of wavelengths with fast pumping/venting speed and possibility of modulating the device temperature (room temperature to ~150deg).

physics.app-ph↗

arXiv2020

Electron confinement by laser-driven azimuthal magnetic fields during direct laser acceleration

A laser-driven azimuthal plasma magnetic field is known to facilitate electron energy gain from the irradiating laser pulse. The enhancement is due to changes in the orientation between the laser electric field and electron velocity caused by magnetic field deflections. Transverse electron confinement is critical for realizing this concept experimentally. We find that the phase velocity of the laser pulse has a profound impact on the transverse size of electron trajectories. The transverse size remains constant below a threshold energy that depends on the degree of the superluminosity and it increases with the electron energy above the threshold. This increase can cause electron losses in tightly focused laser pulses. We show using 3D particle-in-cell simulations that the electron energy gain can be significantly increased by increasing laser power at fixed intensity due to the increased electron confinement. This finding makes a strong case for designing experiments at multi-PW laser facilities.

physics.plasm-ph↗

arXiv2020

Structure-Level Knowledge Distillation For Multilingual Sequence Labeling

Multilingual sequence labeling is a task of predicting label sequences using a single unified model for multiple languages. Compared with relying on multiple monolingual models, using a multilingual model has the benefit of a smaller model size, easier in online serving, and generalizability to low-resource languages. However, current multilingual models still underperform individual monolingual models significantly due to model capacity limitations. In this paper, we propose to reduce the gap between monolingual models and the unified multilingual model by distilling the structural knowledge of several monolingual models (teachers) to the unified multilingual model (student). We propose two novel KD methods based on structure-level information: (1) approximately minimizes the distance between the student's and the teachers' structure level probability distributions, (2) aggregates the structure-level knowledge to local distributions and minimizes the distance between two local probability distributions. Our experiments on 4 multilingual tasks with 25 datasets show that our approaches outperform several strong baselines and have stronger zero-shot generalizability than both the baseline model and teacher models.

cs.CL↗

arXiv2020

Automatic low-bit hybrid quantization of neural networks through meta learning

Model quantization is a widely used technique to compress and accelerate deep neural network (DNN) inference, especially when deploying to edge or IoT devices with limited computation capacity and power consumption budget. The uniform bit width quantization across all the layers is usually sub-optimal and the exploration of hybrid quantization for different layers is vital for efficient deep compression. In this paper, we employ the meta learning method to automatically realize low-bit hybrid quantization of neural networks. A MetaQuantNet, together with a Quantization function, are trained to generate the quantized weights for the target DNN. Then, we apply a genetic algorithm to search the best hybrid quantization policy that meets compression constraints. With the best searched quantization policy, we subsequently retrain or finetune to further improve the performance of the quantized target network. Extensive experiments demonstrate the performance of searched hybrid quantization scheme surpass that of uniform bitwidth counterpart. Compared to the existing reinforcement learning (RL) based hybrid quantization search approach that relies on tedious explorations, our meta learning approach is more efficient and effective for any compression requirements since the MetaQuantNet only needs be trained once.

cs.LG↗

arXiv2020

Lyapunov-type Conditions for Non-strong Ergodicity of Markov Processes

We present Lyapunov-type conditions for non-strong ergodicity of Markov processes. Some concrete models are discussed including diffusion processes on Riemannian manifolds and Ornstein-Uhlenbeck processes driven by symmetric $α$-stable processes. For SDE driven by $α$-stable process ($α\in (0,2]$) with polynomial drift, the strong ergodicity or not is independent on $α$.

math.PR↗

arXiv2020

Stability of Multidimensional Thermoelastic Contact Discontinuities

We study the system of nonisentropic thermoelasticity describing the motion of thermoelastic nonconductors of heat in two and three spatial dimensions, where the frame-indifferent constitutive relation generalizes that for compressible neo-Hookean materials. Thermoelastic contact discontinuities are characteristic discontinuities for which the velocity is continuous across the discontinuity interface. Mathematically, this renders a nonlinear multidimensional hyperbolic problem with a characteristic free boundary. We identify a stability condition on the piecewise constant background states and establish the linear stability of thermoelastic contact discontinuities in the sense that the variable coefficient linearized problem satisfies a priori tame estimates in the usual Sobolev spaces under small perturbations. Our tame estimates for the linearized problem do not break down when the strength of thermoelastic contact discontinuities tends to zero. The missing normal derivatives are recovered from the estimates of several quantities relating to physical involutions. In the estimate of tangential derivatives, there is a significant new difficulty, namely the presence of characteristic variables in the boundary conditions. To overcome this difficulty, we explore an intrinsic cancellation effect, which reduces the boundary terms to an instant integral. Then we can absorb the instant integral into the instant tangential energy by means of the interpolation argument and an explicit estimate for the traces on the hyperplane.

math.AP↗

arXiv2020

Central Similarity Quantization for Efficient Image and Video Retrieval

Existing data-dependent hashing methods usually learn hash functions from pairwise or triplet data relationships, which only capture the data similarity locally, and often suffer from low learning efficiency and low collision rate. In this work, we propose a new \emph{global} similarity metric, termed as \emph{central similarity}, with which the hash codes of similar data pairs are encouraged to approach a common center and those for dissimilar pairs to converge to different centers, to improve hash learning efficiency and retrieval accuracy. We principally formulate the computation of the proposed central similarity metric by introducing a new concept, i.e., \emph{hash center} that refers to a set of data points scattered in the Hamming space with a sufficient mutual distance between each other. We then provide an efficient method to construct well separated hash centers by leveraging the Hadamard matrix and Bernoulli distributions. Finally, we propose the Central Similarity Quantization (CSQ) that optimizes the central similarity between data points w.r.t.\ their hash centers instead of optimizing the local similarity. CSQ is generic and applicable to both image and video hashing scenarios. Extensive experiments on large-scale image and video retrieval tasks demonstrate that CSQ can generate cohesive hash codes for similar data pairs and dispersed hash codes for dissimilar pairs, achieving a noticeable boost in retrieval performance, i.e. 3\%-20\% in mAP over the previous state-of-the-arts. The code is at: \url{https://github.com/yuanli2333/Hadamard-Matrix-for-hashing}

cs.CV↗

arXiv2020

Non-thermal vibrations in biased molecular junctions

We study vibrational statistics in current-carrying model molecular junctions using master equation approach. Especially, we concentrate on the validity of using an effective temperature $T_{\rm eff}$ to characterize the nonequilibrium steady state of a vibrational mode. We identify cases where a single $T_{\rm eff}$ can not fully describe one vibrational state. In such cases, the probability distribution among different vibrational states does not follow the Boltzmann type. Consequently, the actual entropy (free energy) of the vibrational mode is lower (higher) than the corresponding thermal value given by $T_{\rm eff}$, indicating extra work can be extracted from these states. Our results will be useful for the study of non-thermal vibrational state in thermodynamics of nanoscale systems, and its usage in nanoscale heat engines.

cond-mat.mes-hall↗

arXiv2020

SCUBA-2 Ultra Deep Imaging Eao Survey (Studies) III: Multi-wavelength properties, luminosity functions and preliminary source catalog of 450-$μ$m-selected galaxies

We construct a SCUBA-2 450-$μ$m map in the COSMOS field that covers an area of 300 arcmin$^{2}$ and reaches a 1$σ$ noise level of 0.65 mJy in the deepest region. We extract 256 sources detected at 450 $μ$m with signal-to-noise ratio $>$ 4.0 and analyze the physical properties of their multi-wavelength counterparts. We find that most of the sources are at $z\lesssim3$, with a median of $z = 1.79^{+0.03}_{-0.15}$. About $35^{+32}_{-25}$% of our sources are classified as starburst galaxies based on their total star-formation rates (SFRs) and stellar masses ($M_{\ast}$). By fitting the far-infrared spectral energy distributions, we find that our 450-$μ$m-selected sample has a wide range of dust temperatures (20 K $ \lesssim T_{\rm d} \lesssim$ 60 K), with a median of ${T}_{\rm d} = 38.3^{+0.4}_{-0.9}$ K. We do not find a redshift evolution in dust temperature for sources with $L_{\rm IR}$ > $10^{12}$ $\rm L_\odot$ at $z<3$. However, we find a moderate correlation where dust temperature increases with the deviation from the SFR-$M_{\ast}$ relation. The increase in dust temperature also correlates with optical morphology, which is consistent with merger-triggered starbursts in sub-millimeter galaxies. Our galaxies do not show the tight IRX-$β_{\rm UV}$ correlation that has been observed in the local Universe. We construct the infrared luminosity functions of our 450-$μ$m sources and measure their comoving SFR densities. The contribution of the $L_{\rm IR}$ > $10^{12}$ $\rm L_\odot$ population to the SFR density rises dramatically from $z$ = 0 to 2 ($\propto$ ($1+z$)$^{3.9\pm1.1}$) and dominates the total SFR density at $z \gtrsim 2$.

astro-ph.GA↗

arXiv2020

Numerical analysis of two Galerkin discretizations with graded temporal grids for fractional evolution equations

Two numerical methods with graded temporal grids are analyzed for fractional evolution equations. One is a low-order discontinuous Galerkin (DG) discretization in the case of fractional order $0<α<1$, and the other one is a low-order Petrov Galerkin (PG) discretization in the case of fractional order $1<α<2$. By a new duality technique, pointwise-in-time error estimates of first-order and $ (3-α) $-order temporal accuracies are respectively derived for DG and PG, under reasonable regularity assumptions on the initial value. Numerical experiments are performed to verify the theoretical results.

math.NA↗