arXiv subjects

Jun Xu

Publications and source records attributed to Jun Xu.

At least 433 records · Page 24Linked to original sources

arXiv2018

A Tree Search Algorithm for Sequence Labeling

In this paper we propose a novel reinforcement learning based model for sequence tagging, referred to as MM-Tag. Inspired by the success and methodology of the AlphaGo Zero, MM-Tag formalizes the problem of sequence tagging with a Monte Carlo tree search (MCTS) enhanced Markov decision process (MDP) model, in which the time steps correspond to the positions of words in a sentence from left to right, and each action corresponds to assign a tag to a word. Two long short-term memory networks (LSTM) are used to summarize the past tag assignments and words in the sentence. Based on the outputs of LSTMs, the policy for guiding the tag assignment and the value for predicting the whole tagging accuracy of the whole sentence are produced. The policy and value are then strengthened with MCTS, which takes the produced raw policy and value as inputs, simulates and evaluates the possible tag assignments at the subsequent positions, and outputs a better search policy for assigning tags. A reinforcement learning algorithm is proposed to train the model parameters. Our work is the first to apply the MCTS enhanced MDP model to the sequence tagging task. We show that MM-Tag can accurately predict the tags thanks to the exploratory decision making mechanism introduced by MCTS. Experimental results show based on a chunking benchmark showed that MM-Tag outperformed the state-of-the-art sequence tagging baselines including CRF and CRF with LSTM.

cs.CL↗

arXiv2018

Modeling Diverse Relevance Patterns in Ad-hoc Retrieval

Assessing relevance between a query and a document is challenging in ad-hoc retrieval due to its diverse patterns, i.e., a document could be relevant to a query as a whole or partially as long as it provides sufficient information for users' need. Such diverse relevance patterns require an ideal retrieval model to be able to assess relevance in the right granularity adaptively. Unfortunately, most existing retrieval models compute relevance at a single granularity, either document-wide or passage-level, or use fixed combination strategy, restricting their ability in capturing diverse relevance patterns. In this work, we propose a data-driven method to allow relevance signals at different granularities to compete with each other for final relevance assessment. Specifically, we propose a HIerarchical Neural maTching model (HiNT) which consists of two stacked components, namely local matching layer and global decision layer. The local matching layer focuses on producing a set of local relevance signals by modeling the semantic matching between a query and each passage of a document. The global decision layer accumulates local signals into different granularities and allows them to compete with each other to decide the final relevance score. Experimental results demonstrate that our HiNT model outperforms existing state-of-the-art retrieval models significantly on benchmark ad-hoc retrieval datasets.

cs.IR↗

arXiv2018

Combined Constraints on the Equation of State of Dense Neutron-Rich Matter from Terrestrial Experiments and Observations of Neutron Stars

Within the parameter space of equation of state (EOS) of dense neutron-rich matter limited by existing constraints mainly from terrestrial nuclear experiments, we investigate how the neutron star maximum mass $M_{\rm{max}}>2.01\pm0.04$ M$_\odot$, radius $10.62<R_{\rm{1.4}}< 12.83$ km and tidal deformability $Λ_{1.4}\leq800$ of canonical neutron stars all together constrain the EOS of dense neutron-rich nucleonic matter. While the 3-D parameter space of $K_{\rm{sym}}$ (curvature of nuclear symmetry energy), $J_{\rm{sym}}$ and $J_0$ (skewness of the symmetry energy and EOS of symmetric nuclear matter, respectively) are narrowed down significantly by the observational constraints, more data are needed to pin down the individual values of $K_{\rm{sym}}$, $J_{\rm{sym}}$ and $J_0$ with quantified uncertainties. The $J_0$ largely controls the maximum mass of neutron stars. While the EOS with $J_0=0$ is sufficiently stiff to support neutron stars as massive as 2.37 M$_{\odot}$, to support the hyperthetical ones as massive as 2.74 M$_{\odot}$ (composite mass of GW170817) requires $J_0$ to be larger than its currently known maximum value of about 400 MeV and beyond the causality limit. The upper limit on the tidal deformability of $Λ_{1.4}=800$ from the recent observation of GW170817 is found to provide upper limits on some EOS parameters consistent with but far less restrictive than the existing constraints of other observables studied.

nucl-th↗

arXiv2018

MQGrad: Reinforcement Learning of Gradient Quantization in Parameter Server

One of the most significant bottleneck in training large scale machine learning models on parameter server (PS) is the communication overhead, because it needs to frequently exchange the model gradients between the workers and servers during the training iterations. Gradient quantization has been proposed as an effective approach to reducing the communication volume. One key issue in gradient quantization is setting the number of bits for quantizing the gradients. Small number of bits can significantly reduce the communication overhead while hurts the gradient accuracies, and vise versa. An ideal quantization method would dynamically balance the communication overhead and model accuracy, through adjusting the number bits according to the knowledge learned from the immediate past training iterations. Existing methods, however, quantize the gradients either with fixed number of bits, or with predefined heuristic rules. In this paper we propose a novel adaptive quantization method within the framework of reinforcement learning. The method, referred to as MQGrad, formalizes the selection of quantization bits as actions in a Markov decision process (MDP) where the MDP states records the information collected from the past optimization iterations (e.g., the sequence of the loss function values). During the training iterations of a machine learning algorithm, MQGrad continuously updates the MDP state according to the changes of the loss function. Based on the information, MDP learns to select the optimal actions (number of bits) to quantize the gradients. Experimental results based on a benchmark dataset showed that MQGrad can accelerate the learning of a large scale deep neural network while keeping its prediction accuracies.

cs.LG↗

arXiv2018

Comparison of heavy-ion transport simulations: Collision integral in a box

Simulations by transport codes are indispensable to extract valuable physics information from heavy ion collisions. In order to understand the origins of discrepancies between different widely used transport codes, we compare 15 such codes under controlled conditions of a system confined to a box with periodic boundary, initialized with Fermi-Dirac distributions at saturation density and temperatures of either 0 or 5 MeV. In such calculations, one is able to check separately the different ingredients of a transport code. In this second publication of the code evaluation project, we only consider the two-body collision term, i.e. we perform cascade calculations. When the Pauli blocking is artificially suppressed, the collision rates are found to be consistent for most codes (to within $1\%$ or better) with analytical results, or completely controlled results of a basic cascade code after eliminating the correlations within the same pair of colliding particles. In calculations with active Pauli blocking, the blocking probability was found to deviate from the expected reference values. The reason is found in substantial phase-space fluctuations and smearing tied to numerical algorithms and model assumptions in the representation of phase space. This results in the reduction of the blocking probability in most transport codes, so that the simulated system gradually evolves away from the Fermi-Dirac towards a Boltzmann distribution. As a result of this investigation, we are able to make judgements about the most effective strategies in transport simulations for determining the collision probabilities and the Pauli blocking. Investigation in a similar vein of other ingredients in transport calculations, like the mean field propagation or the production of nucleon resonances and mesons, will be discussed in the future publications.

nucl-th↗

arXiv2018

Nonlinear Unknown Input and State Estimation Algorithm in Mobile Robots

This technical report provides the description and the derivation of a novel nonlinear unknown input and state estimation algorithm (NUISE) for mobile robots. The algorithm is designed for real-world robots with nonlinear dynamic models and subject to stochastic noises on sensing and actuation. Leveraging sensor readings and planned control commands, the algorithm detects and quantifies anomalies on both sensors and actuators. Later, we elaborate the dynamic models of two distinctive mobile robots for the purpose of demonstrating the application of NUISE. This report serves as a supplementary document for [1].

eess.SY↗

arXiv2018

Real-world Noisy Image Denoising: A New Benchmark

Most of previous image denoising methods focus on additive white Gaussian noise (AWGN). However,the real-world noisy image denoising problem with the advancing of the computer vision techiniques. In order to promote the study on this problem while implementing the concurrent real-world image denoising datasets, we construct a new benchmark dataset which contains comprehensive real-world noisy images of different natural scenes. These images are captured by different cameras under different camera settings. We evaluate the different denoising methods on our new dataset as well as previous datasets. Extensive experimental results demonstrate that the recently proposed methods designed specifically for realistic noise removal based on sparse or low rank theories achieve better denoising performance and are more robust than other competing methods, and the newly proposed dataset is more challenging. The constructed dataset of real photographs is publicly available at \url{https://github.com/csjunxu/PolyUDataset} for researchers to investigate new real-world image denoising methods. We will add more analysis on the noise statistics in the real photographs of our new dataset in the next version of this article.

cs.CV↗

arXiv2018

Imaging nanoscale spatial modulation of a relativistic electron beam with a MeV ultrafast electron microscope

Accelerator-based MeV ultrafast electron microscope (MUEM) has been proposed as a promising tool to study structural dynamics at the nanometer spatial scale and picosecond temporal scale. Here we report experimental tests of a prototype MUEM where high quality images with nanoscale fine structures were recorded with a pulsed 3 MeV picosecond electron beam. The temporal and spatial resolution of the MUEM operating in single-shot mode is about 4 ps (FWHM) and 100 nm (FWHM), corresponding to a temporal-spatial resolution of 4e-19 s*m, about 2 orders of magnitude higher than that achieved with state-of-the-art single-shot keV UEM. Using this instrument we offer the demonstration of visualizing the nanoscale periodic spatial modulation of an electron beam, which may be converted into longitudinal density modulation through emittance exchange to enable production of high-power coherent radiation at short wavelengths. Our results mark a great step towards single-shot nanometer-resolution MUEMs and compact intense x-ray sources that may have wide applications in many areas of science.

physics.acc-ph↗

arXiv2018

Reducing detrimental electrostatic effects in Casimir-force measurements and Casimir-force-based microdevices

It is well known that residual electrostatic forces create significant difficulties to precise measurements of the Casimir force and to wide use of Casimir-operated microdevices. We experimentally demonstrate that with the help of Ar-ion cleaning of the surfaces it is possible to make electrostatic effects negligibly small as compared to the Casimir interaction. Our experimental setup consists of the dynamic atomic force microscope supplemented with an Ar-ion gun and argon reservoir. The residual potential difference between the Au-coated surfaces of a sphere and a plate was measured both before and after the in situ Ar-ion cleaning. It is shown that this cleaning decreases the magnitude of the residual potential by up to an order of magnitude and makes it almost independent on separation. The gradient of the Casimir force was measured using ordinary samples subjected to the Ar-ion cleaning. The obtained results are shown to be in good agreement with both previous precision measurements using the specially selected samples and with theoretical predictions of the Lifshitz theory. The conclusion is made that the suggested method of in situ Ar-ion cleaning is effective in reducing the electrostatic effects and therefore is a great resource for experiments on measuring the Casimir interaction and for Casimir-operated microdevices.

quant-ph↗

arXiv2018

Optimize transfer learning for lung diseases in bronchoscopy using a new concept: sequential fine-tuning

Bronchoscopy inspection as a follow-up procedure from the radiological imaging plays a key role in lung disease diagnosis and determining treatment plans for the patients. Doctors needs to make a decision whether to biopsy the patients timely when performing bronchoscopy. However, the doctors also needs to be very selective with biopsies as biopsies may cause uncontrollable bleeding of the lung tissue which is life-threaten. To help doctors to be more selective on biopsies and provide a second opinion on diagnosis, in this work, we propose a computer-aided diagnosis (CAD) system for lung diseases including cancers and tuberculosis (TB). The system is developed based on transfer learning. We propose a novel transfer learning method: sentential fine-tuning . Compared to traditional fine-tuning methods, our methods achieves the best performance. We obtained a overall accuracy of 77.0% a dataset of 81 normal cases, 76 tuberculosis cases and 277 lung cancer cases while the other traditional transfer learning methods achieve an accuracy of 73% and 68%. . The detection accuracy of our method for cancers, TB and normal cases are 87%, 54% and 91% respectively. This indicates that the CAD system has potential to improve lung disease diagnosis accuracy in bronchoscopy and it also might be used to be more selective with biopsies.

cs.CV↗

arXiv2018

How tightly is nuclear symmetry energy constrained by unitary Fermi gas?

We examine critically how tightly the density dependence of nuclear symmetry energy \esym is constrained by the universal equation of state (EOS) of the unitary Fermi gas $E_{\rm{UG}}(ρ)$ considering currently known uncertainties of higher order parameters describing the density dependence of the Equation of State of isospin-asymmetric nuclear matter. We found that $E_{\rm{UG}}(ρ)$ does provide a useful lower boundary for the \esym. However, it does not tightly constrain the correlation between the magnitude $E_{\rm{sym}}(ρ_0)$ and slope $L$ unless the curvature $K_{\rm{sym}}$ of the symmetry energy at saturation density $ρ_0$ is more precisely known. The large uncertainty in the skewness parameters affects the $E_{\rm{sym}}(ρ_0)$ versus $L$ correlation by the same almost as significantly as the uncertainty in $K_{\rm{sym}}$.

nucl-th↗

arXiv2018

Composing Method for the Two-dimensional Scanning Spectra Observed by the New Vacuum Solar Telescope

In this paper we illustrate the technique used by the New Vacuum Solar Telescope to increase the spatial resolution of two-dimensional (2D) solar spectroscopy observation involving two dimensions of space and one of wavelength. Without an image stabilizer at the NVST, a large scale wobble motion is present during the spatial scanning, whose instantaneous amplitude could reach up to 1.3" due to the earth's atmosphere and the precision of the telescope guiding system, and seriously decreases the spatial resolution of 2D spatial maps composed with the scanning spectra. We make the following effort to resolve this problem: the imaging system (e.g., the TiO-band) is used to record and detect the displacement vectors of solar image motion during the raster scan, in both the slit and scanning directions. The spectral data (e.g., the Ha line) which are originally obtained in time sequence are corrected and re-arranged in space according to those displacement vectors. Raster scans are carried out in several active regions with different seeing conditions (two rasters are illustrated in this paper). Given a certain spatial sample and temporal resolution, the spatial resolution of the composed 2D map could be close to that of the slit-jaw image. The resulting quality after correction is quantitatively evaluated with two methods. Two-dimensional physical quantity, such as the line-of-sight velocities in multi-layer of the solar atmosphere, is also inferred demonstrating the effect of this technique.

astro-ph.IM↗

arXiv2018

Nucleon Effective Masses in Neutron-Rich Matter

Various kinds of isovector nucleon effective masses are used in the literature to characterize the momentum/energy dependence of the nucleon symmetry potential or self-energy due to the space/time non-locality of the underlying isovector strong interaction in neutron-rich nucleonic matter. The multifaceted studies on nucleon isovector effective masses are multi-disciplinary in nature. Besides structures, masses and low-lying excited states of nuclei as well as nuclear reactions, studies of the isospin dependence of short-range correlations in nuclei from scatterings of high-energy electrons and protons on heavy nuclei also help understand nucleon effective masses especially the so-called E-mass in neutron-rich matter. A thorough understanding of all kinds of nucleon effective masses has multiple impacts on many interesting issues in both nuclear physics and astrophysics. We review some of the significant progresses made in recent years by the nuclear physics community in resolving some of the hotly debated and longstanding issues regarding nucleon effective masses especially in dense neutron-rich matter.

nucl-th↗

arXiv2017

Better Algorithms for Hybrid Circuit and Packet Switching in Data Centers

Hybrid circuit and packet switching for data center networking (DCN) has received considerable research attention recently. A hybrid-switched DCN employs a much faster circuit switch that is reconfigurable with a nontrivial cost, and a much slower packet switch that is reconfigurable with no cost, to interconnect its racks of servers. The research problem is, given a traffic demand matrix (between the racks), how to compute a good circuit switch configuration schedule so that the vast majority of the traffic demand is removed by the circuit switch, leaving a remaining demand matrix that contains only small elements for the packet switch to handle. In this paper, we propose two new hybrid switch scheduling algorithms under two different scheduling constraints. Our first algorithm, called 2-hop Eclipse, strikes a much better tradeoff between the resulting performance (of the hybrid switch) and the computational complexity (of the algorithm) than the state of the art solution Eclipse/Eclipse++. Our second algorithm, called BFF (best first fit), is the first hybrid switching solution that exploits the potential partial reconfiguration capability of the circuit switch for performance gains.

cs.NI↗

arXiv2017

Locally Smoothed Neural Networks

Convolutional Neural Networks (CNN) and the locally connected layer are limited in capturing the importance and relations of different local receptive fields, which are often crucial for tasks such as face verification, visual question answering, and word sequence prediction. To tackle the issue, we propose a novel locally smoothed neural network (LSNN) in this paper. The main idea is to represent the weight matrix of the locally connected layer as the product of the kernel and the smoother, where the kernel is shared over different local receptive fields, and the smoother is for determining the importance and relations of different local receptive fields. Specifically, a multi-variate Gaussian function is utilized to generate the smoother, for modeling the location relations among different local receptive fields. Furthermore, the content information can also be leveraged by setting the mean and precision of the Gaussian function according to the content. Experiments on some variant of MNIST clearly show our advantages over CNN and locally connected layer.

cs.LG↗

arXiv2017

Revisit of directed flow in relativistic heavy-ion collisions from a multiphase transport model

We have revisited several interesting questions on how the rapidity-odd directed flow is developed in relativistic $^{197}$Au+$^{197}$Au collisions at $\sqrt{s_{NN}}$ = 200 and 39 GeV based on a multiphase transport model. As the partonic phase evolves with time, the slope of the parton directed flow at midrapidity region changes from negative to positive as a result of the later dynamics at 200 GeV, while it remains negative at 39 GeV due to the shorter life time of the partonic phase. The directed flow splitting for various quark species due to their different initial eccentricities is observed at 39 GeV, while the splitting is very small at 200 GeV. From a dynamical coalescence algorithm with Wigner functions, we found that the directed flow of hadrons is a result of competition between the coalescence in momentum and coordinate space as well as further modifications by the hadronic rescatterings.

nucl-th↗

arXiv2017

Learning Visual Features from Snapshots for Web Search

When applying learning to rank algorithms to Web search, a large number of features are usually designed to capture the relevance signals. Most of these features are computed based on the extracted textual elements, link analysis, and user logs. However, Web pages are not solely linked texts, but have structured layout organizing a large variety of elements in different styles. Such layout itself can convey useful visual information, indicating the relevance of a Web page. For example, the query-independent layout (i.e., raw page layout) can help identify the page quality, while the query-dependent layout (i.e., page rendered with matched query words) can further tell rich structural information (e.g., size, position and proximity) of the matching signals. However, such visual information of layout has been seldom utilized in Web search in the past. In this work, we propose to learn rich visual features automatically from the layout of Web pages (i.e., Web page snapshots) for relevance ranking. Both query-independent and query-dependent snapshots are considered as the new inputs. We then propose a novel visual perception model inspired by human's visual search behaviors on page viewing to extract the visual features. This model can be learned end-to-end together with traditional human-crafted features. We also show that such visual features can be efficiently acquired in the online setting with an extended inverted indexing scheme. Experiments on benchmark collections demonstrate that learning visual features from Web page snapshots can significantly improve the performance of relevance ranking in ad-hoc Web retrieval tasks.

cs.IR↗

arXiv2017

Simulating spin dynamics with spin-dependent cross sections in heavy-ion collisions

We have incorporated the spin-dependent nucleon-nucleon cross sections into a Boltzmann-Uehling-Uhlenbeck transport model for the first time, using the spin-singlet and spin-triplet nucleon-nucleon elastic scattering cross sections extracted from the phase-shift analyses of nucleon-nucleon scatterings in free space. We found that the spin splitting of the collective flows is not affected by the spin-dependent cross sections, justifying it as a good probe of the in-medium nuclear spin-orbit interaction. With the in-medium nuclear spin-orbit mean-field potential that leads to local spin polarization, we found that the spin-averaged observables, such as elliptic flows of free nucleons and light clusters, becomes smaller with the spin-dependent differential nucleon-nucleon scattering cross sections.

nucl-th↗