arXiv subjects

Guoqiang Zhang

Publications and source records attributed to Guoqiang Zhang.

At least 73 records · Page 4Linked to original sources

arXiv2020

DS-UI: Dual-Supervised Mixture of Gaussian Mixture Models for Uncertainty Inference

This paper proposes a dual-supervised uncertainty inference (DS-UI) framework for improving Bayesian estimation-based uncertainty inference (UI) in deep neural network (DNN)-based image recognition. In the DS-UI, we combine the classifier of a DNN, i.e., the last fully-connected (FC) layer, with a mixture of Gaussian mixture models (MoGMM) to obtain an MoGMM-FC layer. Unlike existing UI methods for DNNs, which only calculate the means or modes of the DNN outputs' distributions, the proposed MoGMM-FC layer acts as a probabilistic interpreter for the features that are inputs of the classifier to directly calculate the probability density of them for the DS-UI. In addition, we propose a dual-supervised stochastic gradient-based variational Bayes (DS-SGVB) algorithm for the MoGMM-FC layer optimization. Unlike conventional SGVB and optimization algorithms in other UI methods, the DS-SGVB not only models the samples in the specific class for each Gaussian mixture model (GMM) in the MoGMM, but also considers the negative samples from other classes for the GMM to reduce the intra-class distances and enlarge the inter-class margins simultaneously for enhancing the learning ability of the MoGMM-FC layer in the DS-UI. Experimental results show the DS-UI outperforms the state-of-the-art UI methods in misclassification detection. We further evaluate the DS-UI in open-set out-of-domain/-distribution detection and find statistically significant improvements. Visualizations of the feature spaces demonstrate the superiority of the DS-UI.

cs.LG↗

arXiv2020

Measurements of D-D fusion neutrons generated in nanowire array laser plasma using Timepix3 detector

We present the results of neutron detection in a laser plasma experiment with a CD$_2$ nanowire target. A hybrid semiconductor pixel detector Timepix3 covered with neutron converters was used for the detection of neutrons. D-D fusion neutrons were detected in a polyethylene converter through recoiled protons. Both the energy of recoiled protons and the time-of-flight of neutrons (and thus their energy) were determined. We report $(2.4 \pm 1.8) \times 10^7$ neutrons generated for 1~J of incoming laser energy. Furthermore, we proved that Timepix3 is suitable for difficult operational conditions in laser experiments.

physics.ins-det↗

arXiv2020

Near-field cavity optomechanical coupling in a compound semiconductor nanowire

A III-V compound semiconductor nanowire is an attractive material for a novel hybrid quantum interface that interconnects photons, electrons, and phonons through a wavelength-tunable quantum structure embedded in its free-standing structure. In such a nanomechanical element, however, a challenge is how to detect and manipulate a small number of phonons via its tiny mechanical motion. A solution would be to couple an optical cavity to a nanowire by introducing the ``cavity optomechanics'' framework, but the typical size difference between them becomes a barrier to achieving this. Here, we demonstrate near-field coupling of a silica microsphere cavity and an epitaxially grown InP/InAs free-standing nanowire. The evanescent optomechanical coupling enables not only fine probing of the mechanical motion by balanced homodyne interferometry but also tuning of the resonance frequency, linewidth, Duffing nonlinearity, and vibration axis in the nanowire. Combining this cavity optomechanics with epitaxial nanowire engineering opens the way to novel quantum metrology and information processing.

cond-mat.mes-hall↗

arXiv2020

Approximated Orthonormal Normalisation in Training Neural Networks

Generalisation of a deep neural network (DNN) is one major concern when employing the deep learning approach for solving practical problems. In this paper we propose a new technique, named approximated orthonormal normalisation (AON), to improve the generalisation capacity of a DNN model. Considering a weight matrix W from a particular neural layer in the model, our objective is to design a function h(W) such that its row vectors are approximately orthogonal to each other while allowing the DNN model to fit the training data sufficiently accurate. By doing so, it would avoid co-adaptation among neurons of the same layer to be able to improve network-generalisation capacity. Specifically, at each iteration, we first approximate (WW^T)^(-1/2) using its Taylor expansion before multiplying the matrix W. After that, the matrix product is then normalised by applying the spectral normalisation (SN) technique to obtain h(W). Conceptually speaking, AON is designed to turn orthonormal regularisation into orthonormal normalisation to avoid manual balancing the original and penalty functions. Experimental results show that AON yields promising validation performance compared to orthonormal regularisation.

cs.LG↗

arXiv2019

A unified inverse scattering transform and soliton solutions of the nonlocal modified KdV equation with non-zero boundary conditions

We present a rigorous theory of a unified and simple inverse scattering transform (IST) for both focusing and defocusing real nonlocal (reverse-space-time) modified Korteweg-de Vries (mKdV) equations with non-zero boundary conditions (NZBCs) at infinity. The IST problems for the nonlocal equations with NZBCs are more complicated then ones for the local equations with NZBCs. The suitable uniformization variable is introduced in order to make the direct and inverse problems be established on a complex plane instead of a two-sheeted Riemann surface. The direct scattering problem establishes the analyticity, symmetries, and asymptotic behaviors of Jost solutions and scattering matrix, and properties of discrete spectra. The inverse problem is formulated and solved by means of a matrix-valued Riemann-Hilbert problem. The reconstruction formula, trace formulae, and theta conditions are obtained. Finally, the dynamical behaviors of solitons for four different cases for the reflectionless potentials for both focusing and defocusing nonlocal mKdV equations with NZBCs are analyzed in detail.

nlin.SI↗

arXiv2019

Inverse scattering transforms for the focusing and defocusing mKdV equations with nonzero boundary conditions

We explore systematically a rigorous theory of the inverse scattering transforms with matrix Riemann-Hilbert problems for both focusing and defocusing modified Korteweg-de Vries (mKdV) equations with non-zero boundary conditions (NZBCs) at infinity. Using a suitable uniformization variable, the direct and inverse scattering problems are proposed on a complex plane instead of a two-sheeted Riemann surface. For the direct scattering problem, the analyticities, symmetries and asymptotic behaviors of the Jost solutions and scattering matrix, and discrete spectra are established. The inverse problems are formulated and solved with the aid of the Riemann-Hilbert problems, and the reconstruction formulae, trace formulae, and theta conditions are also found. In particular, we present the general solutions for the focusing mKdV equation with NZBCs and both simple and double poles, and for the defocusing mKdV equation with NZBCs and simple poles. Finally, some representative reflectionless potentials are in detail studied to illustrate distinct wave structures for both focusing and defocusing mKdV equations with NZBCs.

nlin.SI↗

arXiv2019

Rapidly Adapting Moment Estimation

Adaptive gradient methods such as Adam have been shown to be very effective for training deep neural networks (DNNs) by tracking the second moment of gradients to compute the individual learning rates. Differently from existing methods, we make use of the most recent first moment of gradients to compute the individual learning rates per iteration. The motivation behind it is that the dynamic variation of the first moment of gradients may provide useful information to obtain the learning rates. We refer to the new method as the rapidly adapting moment estimation (RAME). The theoretical convergence of deterministic RAME is studied by using an analysis similar to the one used in [1] for Adam. Experimental results for training a number of DNNs show promising performance of RAME w.r.t. the convergence speed and generalization performance compared to the stochastic heavy-ball (SHB) method, Adam, and RMSprop.

cs.LG↗

arXiv2019

Object Detection and 3D Estimation via an FMCW Radar Using a Fully Convolutional Network

This paper considers object detection and 3D estimation using an FMCW radar. The state-of-the-art deep learning framework is employed instead of using traditional signal processing. In preparing the radar training data, the ground truth of an object orientation in 3D space is provided by conducting image analysis, of which the images are obtained through a coupled camera to the radar device. To ensure successful training of a fully convolutional network (FCN), we propose a normalization method, which is found to be essential to be applied to the radar signal before feeding into the neural network. The system after proper training is able to first detect the presence of an object in an environment. If it does, the system then further produces an estimation of its 3D position. Experimental results show that the proposed system can be successfully trained and employed for detecting a car and further estimating its 3D position in a noisy environment.

cs.CV↗

arXiv2018

Inverse scattering transforms and N-double-pole solutions for the derivative NLS equation with zero/non-zero boundary conditions

We systematically report a rigorous theory of the inverse scattering transforms (ISTs) for the derivative nonlinear Schrodinger (DNLS) equation with both zero boundary condition (ZBC)/non-zero boundary conditions (NZBCs) at infinity and double poles of analytical scattering coefficients. The scattering theories for both ZBC and NZBCs are addressed. The direct problem establishes the analyticity, symmetries and asymptotic behavior of the Jost solutions and scattering matrix, and properties of discrete spectra. The inverse problems are formulated and solved with the aid of the matrix Riemann-Hilbert problems, and the reconstruction formulae, trace formulae and theta conditions are also posed. In particular, the IST with NZBCs at infinity is proposed by a suitable uniformization variable, which allows the scattering problem to be solved on a standard complex plane instead of a two-sheeted Riemann surface. The reflectionless potentials with double poles for the ZBC and NZBCs are both carried out explicitly by means of determinants. Some representative semi-rational bright-bright soliton, dark-bright soliton, and breather-breather solutions are examined in detail. These results will be useful to further explore and apply the related nonlinear wave phenomena.

nlin.SI↗

arXiv2018

BALSON: Bayesian Least Squares Optimization with Nonnegative L1-Norm Constraint

A Bayesian approach termed BAyesian Least Squares Optimization with Nonnegative L1-norm constraint (BALSON) is proposed. The error distribution of data fitting is described by Gaussian likelihood. The parameter distribution is assumed to be a Dirichlet distribution. With the Bayes rule, searching for the optimal parameters is equivalent to finding the mode of the posterior distribution. In order to explicitly characterize the nonnegative L1-norm constraint of the parameters, we further approximate the true posterior distribution by a Dirichlet distribution. We estimate the statistics of the approximating Dirichlet posterior distribution by sampling methods. Four sampling methods have been introduced. With the estimated posterior distributions, the original parameters can be effectively reconstructed in polynomial fitting problems, and the BALSON framework is found to perform better than conventional methods.

cs.LG↗

arXiv2017

Demonstration of Laser-produced Neutron Diagnostic by Radiative Capture Gamma-rays

We report a new scenario of time-of-flight (TOF) technique in which fast neutrons and delayed gamma-ray signals were both recorded in a millisecond time window in harsh environments induced by high-intensity lasers. The delayed gamma signals, arriving far later than the original fast neutron and often being ignored previously, were identified to be the results of radiative captures of thermalized neutrons. The linear correlation between gamma photon number and the fast neutron yield shows that these delayed gamma events can be employed for neutron diagnosis. This method can reduce the detecting efficiency dropping problem caused by prompt high-flux gamma radiation, and provides a new way for neutron diagnosing in high-intensity laser-target interaction experiments.

physics.ins-det↗

arXiv2017

On Relationship between Primal-Dual Method of Multipliers and Kalman Filter

Recently the primal-dual method of multipliers (PDMM), a novel distributed optimization method, was proposed for solving a general class of decomposable convex optimizations over graphic models. In this work, we first study the convergence properties of PDMM for decomposable quadratic optimizations over tree-structured graphs. We show that with proper parameter selection, PDMM converges to its optimal solution in finite number of iterations. We then apply PDMM for the causal estimation problem over a statistical linear state-space model. We show that PDMM and the Kalman filter have the same update expressions, where PDMM can be interpreted as solving a sequence of quadratic optimizations over a growing chain graph.

math.OC↗

arXiv2017

Training Deep Neural Networks via Optimization Over Graphs

In this work, we propose to train a deep neural network by distributed optimization over a graph. Two nonlinear functions are considered: the rectified linear unit (ReLU) and a linear unit with both lower and upper cutoffs (DCutLU). The problem reformulation over a graph is realized by explicitly representing ReLU or DCutLU using a set of slack variables. We then apply the alternating direction method of multipliers (ADMM) to update the weights of the network layerwise by solving subproblems of the reformulated problem. Empirical results suggest that the ADMM-based method is less sensitive to overfitting than the stochastic gradient descent (SGD) and Adam methods.

cs.LG↗

arXiv2017

Measurements of the Total Cross section for Thermal Neutrons at PNS

In order to measure the total cross section for thermal neutrons, a photoneutron source (PNS, phase 1) has been developed for the acquisition of nuclear data for the Thorium Molten Salt Reactor (TMSR) at the Shanghai Institute of Applied Physics (SINAP). PNS is an electron LINAC pulsed neutron facility that uses the time-of-flight (TOF) technique. It records the neutron TOF and identifies neutrons and $γ$-rays by using a digital signal processing technique. The background is obtained by using a combination of employing 12.8 cm boron-loaded polyethylene(PEB) (5$\%$ w.t.) to block the flight path and Monte Carlo methods. The neutron total cross sections of natural beryllium are measured in the neutron energy region from 0.007 to 0.1 eV. The present measurement result is compared with the fold Harvey data with the response function of PNS.

nucl-ex↗

arXiv2017

Continuous-wave operation and 10-Gb/s direct modulation of InAsP/InP sub-wavelength nanowire laser on silicon photonic crystal

We demonstrated sub-wavelength (~111 nm diameter) single nanowire (NW) continuous wave (CW) lasers on silicon photonic crystal in the telecom-band with direct modulation at 10 Gb/s by optical pumping at cryogenic temperatures. To estimate the small signal response and pseudo-random bit sequence (PRBS) modulation of our CW lasers, we employed a new signal detection technique that employs a superconducting single photon detector and a time-correlated single photon counting module. The results showed that our NW laser was unambiguously modulated at above 10 Gb/s and an open eye pattern was obtained. This is the first demonstration of a telecom-band CW NW laser with high-speed PRBS modulation.

physics.optics↗

arXiv2011

Optimal Index Assignment for Multiple Description Scalar Quantization

We provide a method for designing an optimal index assignment for scalar K-description coding. The method stems from a construction of translated scalar lattices, which provides a performance advantage by exploiting a so-called staggered gain. Interestingly, generation of the optimal index assignment is based on a lattice in K-1 dimensional space. The use of the K-1 dimensional lattice facilitates analytic insight into the performance and eliminates the need for a greedy optimization of the index assignment. It is shown that that the optimal index assignment is not unique. This is illustrated for the two-description case, where a periodic index assignment is selected from possible optimal assignments and described in detail. The new index assignment is applied to design of a K-description quantizer, which is found to outperform a reference K-description quantizer at high rates. The performance advantage due to the staggered gain increases with increasing redundancy among the descriptions.

cs.IT↗

arXiv2010

LANC: locality-aware network coding for better P2P traffic localization

As ISPs begin to cooperate to expose their network locality information as services, e.g., P4P, solutions based on locality information provision for P2P traffic localization will soon approach their capability limits. A natural question is: can we do any better provided that no further locality information improvement can be made? This paper shows how the utility of locality information could be limited by conventional P2P data scheduling algorithms, even as sophisticated as the local rarest first policy. Network coding's simplified data scheduling makes it competent for improving P2P application's throughput. Instead of only using locality information in the topology construction, this paper proposes the locality-aware network coding (LANC) that uses locality information in both the topology construction and downloading decision, and demonstrates its exceptional ability for P2P traffic localization. The randomization introduced by network coding enhances the chance for a peer to find innovative blocks in its neighborhood. Aided by proper locality-awareness, the probability for a peer to get innovative blocks from its proximity will increase as well, resulting in more efficient use of network resources. Extensive simulation results show that LANC can significantly reduce P2P traffic redundancy without sacrificing application-level performance. Aided by the same locality knowledge, the traffic redundancies of LANC in most cases are less than 50\% of the current best approach that does not use network coding.

cs.DC↗

arXiv2010

Phase Changes in the Evolution of the IPv4 and IPv6 AS-Level Internet Topologies

In this paper we investigate the evolution of the IPv4 and IPv6 Internet topologies at the autonomous system (AS) level over a long period of time.We provide abundant empirical evidence that there is a phase transition in the growth trend of the two networks. For the IPv4 network, the phase change occurred in 2001. Before then the network's size grew exponentially, and thereafter it followed a linear growth. Changes are also observed around the same time for the maximum node degree, the average node degree and the average shortest path length. For the IPv6 network, the phase change occurred in late 2006. It is notable that the observed phase transitions in the two networks are different, for example the size of IPv6 network initially grew linearly and then shifted to an exponential growth. Our results show that following decades of rapid expansion up to the beginning of this century, the IPv4 network has now evolved into a mature, steady stage characterised by a relatively slow growth with a stable network structure; whereas the IPv6 network, after a slow startup process, has just taken off to a full speed growth. We also provide insight into the possible impact of IPv6-over-IPv4 tunneling deployment scheme on the evolution of the IPv6 network. The Internet topology generators so far are based on an inexplicit assumption that the evolution of Internet follows non-changing dynamic mechanisms. This assumption, however, is invalidated by our results.Our work reveals insights into the Internet evolution and provides inputs to future AS-Level Internet models.

cs.NI↗