arXiv subjects

Ting Zhang

Publications and source records attributed to Ting Zhang.

At least 217 records · Page 12Linked to original sources

arXiv2018

IGCV$2$: Interleaved Structured Sparse Convolutional Neural Networks

In this paper, we study the problem of designing efficient convolutional neural network architectures with the interest in eliminating the redundancy in convolution kernels. In addition to structured sparse kernels, low-rank kernels and the product of low-rank kernels, the product of structured sparse kernels, which is a framework for interpreting the recently-developed interleaved group convolutions (IGC) and its variants (e.g., Xception), has been attracting increasing interests. Motivated by the observation that the convolutions contained in a group convolution in IGC can be further decomposed in the same manner, we present a modularized building block, {IGCV$2$:} interleaved structured sparse convolutions. It generalizes interleaved group convolutions, which is composed of two structured sparse kernels, to the product of more structured sparse kernels, further eliminating the redundancy. We present the complementary condition and the balance condition to guide the design of structured sparse kernels, obtaining a balance among three aspects: model size, computation complexity and classification accuracy. Experimental results demonstrate the advantage on the balance among these three aspects compared to interleaved group convolutions and Xception, and competitive performance compared to other state-of-the-art architecture design methods.

cs.CV↗

arXiv2018

Matching Natural Language Sentences with Hierarchical Sentence Factorization

Semantic matching of natural language sentences or identifying the relationship between two sentences is a core research problem underlying many natural language tasks. Depending on whether training data is available, prior research has proposed both unsupervised distance-based schemes and supervised deep learning schemes for sentence matching. However, previous approaches either omit or fail to fully utilize the ordered, hierarchical, and flexible structures of language objects, as well as the interactions between them. In this paper, we propose Hierarchical Sentence Factorization---a technique to factorize a sentence into a hierarchical representation, with the components at each different scale reordered into a "predicate-argument" form. The proposed sentence factorization technique leads to the invention of: 1) a new unsupervised distance metric which calculates the semantic distance between a pair of text snippets by solving a penalized optimal transport problem while preserving the logical relationship of words in the reordered sentences, and 2) new multi-scale deep learning models for supervised semantic training, based on factorized sentence hierarchies. We apply our techniques to text-pair similarity estimation and text-pair relationship classification tasks, based on multiple datasets such as STSbenchmark, the Microsoft Research paraphrase identification (MSRP) dataset, the SICK dataset, etc. Extensive experiments show that the proposed hierarchical sentence factorization can be used to significantly improve the performance of existing unsupervised distance-based metrics as well as multiple supervised deep learning models based on the convolutional neural network (CNN) and long short-term memory (LSTM).

cs.CL↗

arXiv2018

Muliti-scale regularity of axisymmetric Navier-Stokes equations

By applying the delicate \textit{a priori} estimates for the equations of $(Φ,Γ)$, which is introduced in the previous work, we obtain some multi-scale regularity criteria of the swirl component $u^θ$ for the 3D axisymmetric Navier-Stokes equations. In particularly, the solution $\mathbf{u}$ can be continued beyond the time $T$, provided that $u^θ$ satiesfies $$ u^θ \in L^{p}_{T}L^{q_{v}}_{v}L^{q_{h},w}_{h},~~\frac{2}{p}+\frac{1}{q_{v}}+\frac{2}{q_{h}}\leq 1, ~2<q_{h}\leq\infty,~\frac{1}{q_{v}}+\frac{2}{q_{h}}<1. $$

math.AP↗

arXiv2018

Magnetic Field Driven Nodal Topological Superconductivity in Monolayer Transition Metal Dichalcogenides

Recently, Ising superconductors which possess in-plane upper critical fields much larger than the Pauli limit field are under intense experimental study. Many monolayer or few layer transition metal dichalcogenides are shown to be Ising superconductors. In this work, we show that in a wide range of experimentally accessible regimes where the in-plane magnetic field is higher than the Pauli limit field but lower than $H_{c2}$, a 2H-structure monolayer NbSe$_2$ or simiarly TaS$_2$ becomes a nodal topological superconductor. The bulk nodal points appear on the $Γ- M$ lines of the Brillouin zone where the Ising SOC vanishes. The nodal points are connected by Majorana flat bands, similar to the Weyl points being connected by surface Fermi arcs in Weyl semimetals. The Majorana flat bands are associated with a large number of zero energy Majorana fermion edge modes which induce spin-triplet Cooper pairs. This work demonstrates an experimentally feasible way to realise Majorana fermions in nodal topological superconductor, without any fining tuning of experimental parameters.

cond-mat.supr-con↗

arXiv2017

Local and global strong solutions to the stochastic incompressible Navier-Stokes equations in critical Besov space

Considering the stochastic Navier-Stokes system in $\mathbb{R}^d$ forced by a multiplicative white noise, we establish the local existence and uniqueness of the strong solution when the initial data take values in the critical space $\dot{B}_{p,r}^{\frac{d}{p}-1}(\mathbb{R}^d)$. The proof is based on the contraction mapping principle, stopping time and stochastic estimates. Then we prove the global existence of strong solutions in probability if the initial data are sufficiently small, which contain a class of highly oscillating "large" data.

math.AP↗

arXiv2017

Composite Quantization

This paper studies the compact coding approach to approximate nearest neighbor search. We introduce a composite quantization framework. It uses the composition of several ($M$) elements, each of which is selected from a different dictionary, to accurately approximate a $D$-dimensional vector, thus yielding accurate search, and represents the data vector by a short code composed of the indices of the selected elements in the corresponding dictionaries. Our key contribution lies in introducing a near-orthogonality constraint, which makes the search efficiency is guaranteed as the cost of the distance computation is reduced to $O(M)$ from $O(D)$ through a distance table lookup scheme. The resulting approach is called near-orthogonal composite quantization. We theoretically justify the equivalence between near-orthogonal composite quantization and minimizing an upper bound of a function formed by jointly considering the quantization error and the search cost according to a generalized triangle inequality. We empirically show the efficacy of the proposed approach over several benchmark datasets. In addition, we demonstrate the superior performances in other three applications: combination with inverted multi-index, quantizing the query for mobile search, and inner-product similarity search.

cs.CV↗

arXiv2017

Berry phase induced localization to anti-localization transition in two-dimensional Dirac fermion systems

We study theoretically the electrical transport of two-dimensional (2D) massive Dirac fermions, which are described by the 2 by 2 massive Dirac Hamiltonian, and with a gap at the charge neutrality point. Through analytical diagrammatical calculations of electrical conductivity in the presence of long range Coulomb scattering centers, we show that attendant with the variation of the Berry phase from 0 to π as the Fermi energy moves away from the Dirac point/band boundary, a continuous Anderson-localization (AL) to weak-localization (WL), and further to weak anti-localization (WAL) transition occurs, implying a change in the sign of the magnetoresistance. Such transition indicates the presence of metal-insulator transition (MIT) in this 2D system, in contrast to the classical scaling theory. The WL to WAL transition occurs at a certain critical Berry phase despite the concentration of Coulomb impurities, while the MIT critical point, which is the distinguishing doping level separating the AL and WL phases, depends on the competition of conventional conductivity and the negative maximally crossed diagram (MCD) corrections near the bottom of conduction band.

cond-mat.mtrl-sci↗

arXiv2017

A concatenating framework of shortcut convolutional neural networks

It is well accepted that convolutional neural networks play an important role in learning excellent features for image classification and recognition. However, in tradition they only allow adjacent layers connected, limiting integration of multi-scale information. To further improve their performance, we present a concatenating framework of shortcut convolutional neural networks. This framework can concatenate multi-scale features by shortcut connections to the fully-connected layer that is directly fed to the output layer. We do a large number of experiments to investigate performance of the shortcut convolutional neural networks on many benchmark visual datasets for different tasks. The datasets include AR, FERET, FaceScrub, CelebA for gender classification, CUReT for texture classification, MNIST for digit recognition, and CIFAR-10 for object recognition. Experimental results show that the shortcut convolutional neural networks can achieve better results than the traditional ones on these tasks, with more stability in different settings of pooling schemes, activation functions, optimizations, initializations, kernel numbers and kernel sizes.

cs.CV↗

arXiv2017

Isolation and Characterization of Few-layer Manganese Thiophosphite

This work reports an experimental study on an antiferromagnetic honeycomb lattice of MnPS$_3$ that couples the valley degree of freedom to a macroscopic antiferromagnetic order. The crystal structure of MnPS$_3$ is identified by high resolution scanning transmission electron microscopy. Layer dependent angle resolved polarized Raman fingerprints of the MnPS$_3$ crystal are obtained and the Raman peak at 383 cm$^{-1}$ exhibits 100% polarity. Temperature dependences of anisotropic magnetic susceptibility of MnPS$_3$ crystal are measured in superconducting quantum interference device. Magnetic parameters like effective magnetic moment, and exchange interaction are extracted from the mean field approximation mode. Ambipolar electronic transport channels in MnPS$_3$ are realized by the liquid gating technique. The conducting channel of MnPS$_3$ offers a unique platform for exploring the spin/valleytronics and magnetic orders in 2D limitation.

cond-mat.mes-hall↗

arXiv2017

Interleaved Group Convolutions for Deep Neural Networks

In this paper, we present a simple and modularized neural network architecture, named interleaved group convolutional neural networks (IGCNets). The main point lies in a novel building block, a pair of two successive interleaved group convolutions: primary group convolution and secondary group convolution. The two group convolutions are complementary: (i) the convolution on each partition in primary group convolution is a spatial convolution, while on each partition in secondary group convolution, the convolution is a point-wise convolution; (ii) the channels in the same secondary partition come from different primary partitions. We discuss one representative advantage: Wider than a regular convolution with the number of parameters and the computation complexity preserved. We also show that regular convolutions, group convolution with summation fusion, and the Xception block are special cases of interleaved group convolutions. Empirical results over standard benchmarks, CIFAR-$10$, CIFAR-$100$, SVHN and ImageNet demonstrate that our networks are more efficient in using parameters and computation complexity with similar or higher accuracy.

cs.CV↗

arXiv2017

A Survey on Learning to Hash

Nearest neighbor search is a problem of finding the data points from the database such that the distances from them to the query point are the smallest. Learning to hash is one of the major solutions to this problem and has been widely studied recently. In this paper, we present a comprehensive survey of the learning to hash algorithms, categorize them according to the manners of preserving the similarities into: pairwise similarity preserving, multiwise similarity preserving, implicit similarity preserving, as well as quantization, and discuss their relations. We separate quantization from pairwise similarity preserving as the objective function is very different though quantization, as we show, can be derived from preserving the pairwise similarities. In addition, we present the evaluation protocols, and the general performance analysis, and point out that the quantization algorithms perform superiorly in terms of search accuracy, search time cost, and space cost. Finally, we introduce a few emerging topics.

cs.CV↗

arXiv2017

Preparation of hierarchical C@MoS2@C sandwiched hollow spheres for Lithium ion batteries

Hierarchical C@MoS2@C hollow spheres with the active MoS2 nanosheets being sandwiched by carbon layers have been produced by means of a modified template method. The process applies polydopamine (PDA) layers which inhibit morphology change of the template thereby enforcing the hollow microsphere structure. In addition, PDA forms complexes with the Mo precursor, leading to an in-situ growth of MoS2 on its surface and preventing the nanosheets from agglomeration. It also supplies the carbon that finally sandwiches the 100-150 nm thin MoS2 spheres. The resulting hierarchically structured material provides a stable microstructure where carbon layers strongly linked to MoS2 offer efficient pathways for electron and ion transfer, and concomitantly buffer the volume changes inevitably appearing during the charge-discharge process. Carbon-sandwiched MoS2-based electrodes exhibit high specific capacity of approximately 900 mA h g-1 after 50 cycles at 0.1 C, excellent cycling stability up to 200 cycles, and superior rate performance. The versatile synthesis method reported here offers a general route to design hollow sandwich structures with a variety of different active materials.

cond-mat.mtrl-sci↗

arXiv2016

Charge Density Wave Phase Transition on the Surface of Electrostatically Doped Multilayer Graphene

We demonstrate that charge density wave (CDW) phase transition occurs on the surface of electronically doped multilayer graphene when the Fermi level approaches the M points (also known as van Hove singularities where the density of states diverge) in the Brillouin zone of graphene band structure. The occurrence of such CDW phase transitions are supported by both the electrical transport measurement and optical measurements in electrostatically doped multilayer graphene. The CDW transition is accompanied with the sudden change of graphene channel resistance at T$_m$= 100K, as well as the splitting of Raman G peak (1580 cm$^{-1}$). The splitting of Raman G peak indicats the lifting of in-plane optical phonon branch degeneracy and the non-degenerate phonon branches are correlated to the lattice reconstructions of graphene -- the CDW phase transition.

cond-mat.mes-hall↗

arXiv2016

Global solutions to the isentropic compressible Navier-Stokes equations with a class of large initial data

In this paper, we consider the global well-posedness problem of the isentropic compressible Navier-Stokes equations in the whole space $\R^N$ with $N\ge2$. In order to better reflect the characteristics of the dispersion equation, we make full use of the role of the frequency on the integrability and regularity of the solution, and prove that the isentropic compressible Navier-Stokes equations admit global solutions when the initial data are close to a stable equilibrium in the sense of suitable hybrid Besov norm. As a consequence, the initial velocity with arbitrary $\dot{B}^{\fr{N}{2}-1}_{2,1}$ norm of potential part $\Pe^\bot u_0$ and large highly oscillating are allowed in our results. The proof relies heavily on the dispersive estimates for the system of acoustics, and a careful study of the nonlinear terms.

math.AP↗

arXiv2016

Deeply-Fused Nets

In this paper, we present a novel deep learning approach, deeply-fused nets. The central idea of our approach is deep fusion, i.e., combine the intermediate representations of base networks, where the fused output serves as the input of the remaining part of each base network, and perform such combinations deeply over several intermediate representations. The resulting deeply fused net enjoys several benefits. First, it is able to learn multi-scale representations as it enjoys the benefits of more base networks, which could form the same fused network, other than the initial group of base networks. Second, in our suggested fused net formed by one deep and one shallow base networks, the flows of the information from the earlier intermediate layer of the deep base network to the output and from the input to the later intermediate layer of the deep base network are both improved. Last, the deep and shallow base networks are jointly learnt and can benefit from each other. More interestingly, the essential depth of a fused net composed from a deep base network and a shallow base network is reduced because the fused net could be composed from a less deep base network, and thus training the fused net is less difficult than training the initial deep base network. Empirical results demonstrate that our approach achieves superior performance over two closely-related methods, ResNet and Highway, and competitive performance compared to the state-of-the-arts.

cs.CV↗

arXiv2016

A unified approach to self-normalized block sampling

The inference procedure for the mean of a stationary time series is usually quite different under various model assumptions because the partial sum process behaves differently depending on whether the time series is short or long-range dependent, or whether it has a light or heavy-tailed marginal distribution. In the current paper, we develop an asymptotic theory for the self-normalized block sampling, and prove that the corresponding block sampling method can provide a unified inference approach for the aforementioned different situations in the sense that it does not require the {\em a priori} estimation of auxiliary parameters. Monte Carlo simulations are presented to illustrate its finite-sample performance. The R function implementing the method is available from the authors.

math.ST↗

arXiv2015

Global axisymmetric solutions of 3D inhomogeneous incompressible Navier-Stokes Systems with nonzero swirl

In this paper, we investigate the global well-posedness for the 3-D inhomogeneous incompressible Navier-Stokes system with the axisymmetric initial data. We prove the global well-posedness provided that $$\|\frac{a_{0}}{r}\|_{\infty} \textrm{ and } \|u_{0}^θ\|_{3} \textrm{ are sufficiently small}. $$ Furthermore, if $\mathbf{u}_0\in L^1$ and $ru^θ_0\in L^1\cap L^2$, we have \begin{equation*} \|u^θ(t)\|_{2}^{2}+\langle t\rangle \|\nabla (u^θ\mathbf{e}_θ)(t)\|_{2}^{2}+t\langle t\rangle(\|u_{t}^θ(t)\|_{2}^{2}+\|Δ(u^θ\mathbf{e}_θ)(t)\|_{2}^{2}) \leq C \langle t\rangle^{-\frac{5}{2}},\ \forall\ t>0. \end{equation*}

math.AP↗

arXiv2015

One-dimensional sawtooth and zigzag lattices for ultracold atoms

We describe tunable optical sawtooth and zigzag lattices for ultracold atoms. Making use of the superlattice generated by commensurate wavelengths of light beams, tunable geometries including zigzag and sawtooth configurations can be realised. We provide an experimentally feasible method to fully control inter- ($t$) and intra- ($t'$) unit-cell tunnelling in zigzag and sawtooth lattices. We analyse the conversion of the lattice geometry from zigzag to sawtooth, and show that a nearly flat band is attainable in the sawtooth configuration by means of tuning the lattice parameters. The bandwidth of the first excited band can be reduced up to 2$\%$ of the ground bandwidth for a wide range of lattice setting. A nearly flat band available in a tunable sawtooth lattice would offer a versatile platform for the study of interaction-driven quantum many-body states with ultracold atoms.

cond-mat.quant-gas↗