Generalized Lanczos method for systematic optimization of tensor network states

We propose a generalized Lanczos method to generate the many-body basis states of quantum lattice models using tensor-network states （TNS）. The ground-state wave function is represented as a linear superposition composed from a set of TNS generated by Lanczos iteration. This method improves significantly the accuracy of the tensor-network algorithm and provides an effective way to enlarge the maximal bond dimension of TNS. The ground state such obtained contains significantly more entanglement than each individual TNS, reproducing correctly the logarithmic size dependence of the entanglement entropy in a critical system. The method can be generalized to non-Hamiltonian systems and to the calculation of low-lying excited states, dynamical correlation functions, and other physical properties of strongly correlated systems.

On the emergence of gravitational dynamics from tensor networks

Tensor networks are used to describe the ground state wavefunction of the quantum many-body system.Recently,it has been shown that a tensor network can generate the anti-de Sitter(AdS)geometry by using the entanglement renormalization approach,which provides a new way to realize bulk reconstruction in the AdS/conformal field theory correspondence.However,whether the dynamical connections can be found between the tensor network and gravity is an important unsolved problem.In this paper,we give a novel proposal to integrate ideas from tensor networks,entanglement entropy,canonical quantization of quantum gravity and the holographic principle and argue that the gravitational dynamics can be generated from a tensor network if the wave function of the latter satisfies the Wheeler–DeWitt equation.

Quantum bit threads of MERA tensor network in large c limit

The Ryu-Takayanagi(RT)formula plays a large role in the current theory of gauge-gravity duality and emergent geometry phenomena.The recent reinterpretation of this formula in terms of a set of"bit threads"is an interesting effort in understanding holography.In this study,we investigate a quantum generalization of the"bit threads"based on a tensor network,with particular focus on the multi-scale entanglement renormalization ansatz(MERA).We demonstrate that,in the large c limit,isometries of the MERA can be regarded as"sources"(or"sinks")of the information flow,which extensively modifies the original picture of bit threads by introducing a new variableρ:density of the isometries.In this modified picture of information flow,the isometries can be viewed as generators of the flow.The strong subadditivity and related properties of the entanglement entropy are also obtained in this new picture.The large c limit implies that classical gravity can emerge from the information flow.

Networked Evolutionary Model of Snow-Drift Game Based on Semi-Tensor Product

This paper investigates the networked evolutionary model based on snow-drift game with the strategy of rewards and penalty. Firstly, by using the semi-tensor product of matrices approach, the mathematical model of the networked evolutionary game is built. Secondly, combined with the matrix expression of logic, the mathematical model is expressed as a dynamic logical system and next converted into its evolutionary dynamic algebraic form. Thirdly, the dynamic evolution process is analyzed and the final level of cooperation is discussed. Finally, the effects of the changes in the rewarding and penalty factors on the level of cooperation in the model are studied separately, and the conclusions are verified by examples.

Criterion of Quantum Entanglement and the Covariance Correlation Tensor in the Theory of Quantum Network

This article discusses the covariance correlation tensor (CCT) in quantum network theory for four Bell bases in detail. Furthermore, it gives the expression of the density operator in terms of CCT for a quantum network of three nodes, thus gives the criterion of entanglement for this case, i.e. the conditions of complete separability and partial separability for a given quantum state of three bodies. Finally it discusses the general case for the quantum network of nodes.

基于TensorFlow的交通标志识别方法研究

交通标志识别系统是智能驾驶系统的重要组成部分;本文分析了现有方法存在的问题,基于TensorFlow框架搭建了改进的卷积神经网络,用于识别交通标志;整个系统在TensorFlow上实现,使用行车记录仪采集的视频验证了本文的算法,结果表明本文算法有一定的实用性,而且在准确率,鲁棒性和实时性等方面也表现较好。

Separability of Pure States and Mixed States of the Quantum Network of Two Nodes

This article discusses the separability of the pure states and mixed states of the quantum network of two nodes by means of the criterion of no entanglement in terms of the covariance correlation tensor in quantum network theory, i.e. for a composite system consisting of two nodes. The covariance correlation tensor is equal to zero for all possible and .

Nodes and layers PageRank centrality for multilayer networks

In this paper, we propose a new centrality algorithm that can simultaneously rank the nodes and layers of multilayer networks, referred to as the MRFNL centrality. The centrality of nodes and layers are obtained by developing a novel iterative algorithm for computing a set of tensor equations. Under some conditions, the existence and uniqueness of this centrality were proven by applying the Brouwer fixed point theorem. Furthermore, the convergence of the proposed iterative algorithm was established. Finally, numerical experiments on a simple multilayer network and two real-world multilayer networks(i.e., Pierre Auger Collaboration and European Air Transportation Networks) are proposed to illustrate the effectiveness of the proposed algorithm and to compare it to other existing centrality measures.

A Matrix Approach to the Modeling and Analysis of Networked Evolutionary Games With Time Delays

Using the semi-tensor product method, this paper investigates the modeling and analysis of networked evolutionary games(NEGs) with finite memories, and presents a number of new results. Firstly, a kind of algebraic expression is formulated for the networked evolutionary games with finite memories, based on which the behavior of the corresponding evolutionary game is analyzed. Secondly, under a proper assumption, the existence of Nash equilibrium of the given networked evolutionary games is proved and a free-type strategy sequence is designed for the convergence to the Nash equilibrium. Finally, an illustrative example is worked out to support the obtained new results.

Context-Aware System Modeling Based on Boolean Control Network

Boolean control network consists of a set of Boolean variables whose state is determined by other variables in the network. Boolean network is used for modeling complex system. In this paper, we have presented a model of a context-aware system used in smart home based on Boolean control networks. This modeling describes the relationship between the context elements (person, time, location, and activity) and services (Morning Call, Sleeping, Guarding, Entertainment, and normal), which is effective to logical inference. We apply semi tensor matrix product to describe the dynamic of the system. This matrix form of expression is a convenient and reasonable way to design logic control system.

基于TT-Tucker分解的无预训练LC卷积神经网络压缩方法

张量训练(TT)分解和Tucker分解是两种有效的卷积神经网络压缩方法。然而,TT和Tucker分解分别面临空间结构信息丢失与计算复杂度高等问题。为解决上述问题,文中考虑了网络结构的信息保留率和资源占用情况,采用学习-压缩(LC)算法的约束型压缩框架,提出了一种基于TT-Tucker分解的无预训练LC卷积神经网络压缩方法(TTLC)。TT-LC方法包括学习步骤和压缩步骤两个部分。学习步骤不需要预训练过程,采用了指数循环学习率方法以提高训练准确率。而在压缩步骤,文中根据TT和Tucker分解的优点以及贝叶斯规则选取全局最优秩的特性,运用经验变分贝叶斯矩阵分解(EVBMF)和贝叶斯优化(BayesOpt)选出合理的秩以指导张量分解,采用TT-LC方法压缩训练后的模型。TT-LC方法既降低了空间结构信息丢失率和计算复杂度,又解决了张量的秩选取不合理导致模型准确率显著下降的问题,可实现模型的双重贝叶斯选秩和双重压缩,获得最优的压缩模型。最后,采用ResNets和VGG网络在CIFAR10与CIFAR100数据集上进行实验。结果表明:对于ResNet32网络,相比于基准方法,文中方法在准确率为92.22%的情况下,获得了69.6%的参数量压缩率和66.7%的浮点计算量压缩率。

融合特征编码和短语交互感知的隐式篇章关系识别

隐式篇章关系识别难度大、普遍性高.从论元编码和论元交互角度入手,提出了一种融合特征编码和短语交互感知的隐式篇章关系识别模型.该模型兼顾了论元本身特征和论元间交互特征的作用,并分别进行了优化.论元编码部分整合了双向长短时记忆网络和循环注意力卷积神经网络,能够更全面地捕获论元全局和局部特征;论元交互部分从短语层级考虑论元间的语义关系建模,构建了短语级交互注意力机制,并利用神经张量网络深入挖掘其中的关系模式,更能体现出论元间潜在的更深层次的关联关系.在宾州篇章树库数据集上的实验结果表明,该模型F1值均优于其他模型.

Anisotropic WM conductivity reconstruction based on diffusion tensor magnetic resonance imaging: a simulation study

The present study aims to estimate the in vivo anisotropic conductivities of the White Matter (WM) tissues by means of Magnetic Resonance Electrical Impedance Tomography (MREIT) technique. The realistic anisotropic volume conductor model with different conductivity properties (scalp, skull, CSF, gray matter and WM) is constructed based on the Diffusion Tensor Magnetic Resonance Imaging (DT- MRI) from a healthy human subject. The Radius Basic Function (RBF)-MREIT algorithm of using only one magnetic flux density component was applied to evaluate the eigenvalues of the anisotropic WM with target values set according to the DT-MRI data based on the Wolter’s model, which is more physiologically reliable. The numerical simulations study performed on the five-layer realistic human head model showed that the conductivity reconstruction method had higher accuracy and better robustness against noise. The pilot research was used to judge the feasibility, meaningfulness and reliability of the MREIT applied on the electrical impedance tomography of the complicated human head tissues including anisotropic characteristics.

基于双通道生成对抗网络的城市用电负荷缺失数据补全方法

用电负荷数据的完整性与有效性在负荷预测等应用中具有重要意义。传统的缺失数据补全方法缺乏对用电负荷和多种外部时空关联信息的挖掘,难以获得高精度的补全结果。文中提出了一种双通道生成对抗网络,对缺失的负荷数据进行补全。首先,根据负荷的周期性变化特征和时空关联性构建三阶负荷张量,并将影响负荷变化的多种外部因素构建为三阶辅助信息张量。然后,为满足两种张量的双输入需求,在生成对抗网络的输入层引入双通道机制,通过卷积与反卷积运算提取张量的特征;为提升网络对张量数据的训练效果和补全精度,将张量分解损失引入原始损失函数,并采用改进的混沌映射粒子群优化算法联合优化超参数和网络。最后,在真实负荷数据集上开展数据补全实验。结果表明,所提方法能够对随机缺失率不超过50%、连续缺失不超过3天的负荷数据进行准确补全。

联合张量补全与循环神经网络的时间序列插补法

现存的插补方法大致分为基于统计的插补法和基于深度学习的插补法。基于统计的插补法只能捕捉线性时间关系,导致无法精准建模时间序列的非线性关系;基于深度学习的插补法往往没有考虑到不同时间序列之间的相关性。针对现有方法的问题,本文提出了联合张量补全与循环神经网络的时间序列插补法。首先,将多元时间序列建模成张量,通过张量的低秩补全捕获不同时间序列之间的关系。其次,提出了一个基于时间的动态权重,将张量插补结果和循环神经网络的预测结果进行融合,避免因为连续缺失导致的预测误差累积。最后,在多个真实的时间序列数据集上对所提方法进行了实验评估,结果显示该模型优于已有相关模型,且基于插补后的时间序列可以提升时间序列预测效果。

Motor imagery training induces changes in brain neural networks in stroke patients

Motor imagery is the mental representation of an action without overt movement or muscle activation. However, the effects of motor imagery on stroke-induced hand dysfunction and brain neural networks are still unknown. We conducted a randomized controlled trial in the China Rehabilitation Research Center. Twenty stroke patients, including 13 males and 7 females, 32–51 years old, were recruited and randomly assigned to the traditional rehabilitation treatment group（PP group, n = 10） or the motor imagery training combined with traditional rehabilitation treatment group（MP group, n = 10）. All patients received rehabilitation training once a day, 45 minutes per session, five times per week, for 4 consecutive weeks. In the MP group, motor imagery training was performed for 45 minutes after traditional rehabilitation training, daily. Action Research Arm Test and the Fugl-Meyer Assessment of the upper extremity were used to evaluate hand functions before and after treatment. Transcranial magnetic stimulation was used to analyze motor evoked potentials in the affected extremity. Diffusion tensor imaging was used to assess changes in brain neural networks. Compared with the PP group, the MP group showed better recovery of hand function, higher amplitude of the motor evoked potential in the abductor pollicis brevis, greater fractional anisotropy of the right dorsal pathway, and an increase in the fractional anisotropy of the bilateral dorsal pathway. Our findings indicate that 4 weeks of motor imagery training combined with traditional rehabilitation treatment improves hand function in stroke patients by enhancing the dorsal pathway. This trial has been registered with the Chinese Clinical Trial Registry（registration number： Chi CTR-OCH-12002238）.

基于U⁃Rnet的重力全张量梯度数据反演

重力反演是通过地表信息获取地下地质体空间结构与物理性质的重要手段之一。每个重力梯度分量反映不同的地质体信息,联合重力梯度分量进行重力反演能够更好地研究地下密度异常体的形态和分布。为此,提出基于神经网络的重力全张量梯度数据反演算法,将U⁃Rnet网络应用于重力全张量数据的三维反演问题。为了检验该算法的有效性,采用六种典型模型进行模拟实验,获得了具有清晰边界和稀疏的反演结果。首先,对比L2和Tversky两种损失函数的反演结果,后者的反演结果能更清晰地反映模型的边界位置;然后,对不同梯度张量组合进行反演,四组实验结果在三个方向(x、y、z)上具有不同的反演精度,组合四的误差最低;最后,将该方法应用于美国德克萨斯州文顿盐丘的FTG数据,反演结果与实际地质信息基本吻合。

基于轻量级全连接张量映射网络的高光谱图像分类方法

近年来,基于卷积神经网络的深度学习模型已经在高光谱图像分类领域取得优异表现.然而,模型性能的提升通常依赖于更深、更宽的网络结构,导致参数量和计算量增长,从而限制了模型在机载或星载载荷中的实际部署.为此,本文提出基于轻量级全连接张量映射网络的高光谱图像分类方法.根据全连接张量网络分解的映射思想以及高光谱图像“图谱合一”的结构特点,本文设计两种张量映射卷积单元,通过使用多个具有全连接结构的小尺寸卷积核代替原始卷积核,降低了卷积层的时间和空间复杂度.此外,基于新单元构建残差双分支张量模块.双分支结构共享同一组权重参数,并采用通道分割操作减少特征通道数,提升特征提取过程的实时性.本文所提模型通过使用新单元和新模块充分挖掘高光谱图像的局部空谱信息和全局光谱信息,有效提高了分类性能并减少硬件资源消耗.在三个常用高光谱图像数据集上的实验结果表明,所提模型相较于其他现有工作具有更高的分类性能以及更低的参数量和计算量.

知识图谱补全方法研究综述

知识图谱是用来描述世界中存在的各种实体和概念以及他们之间的关系的一种语义网络,近年来被广泛应用于智能问答、智能推荐和信息检索等领域。目前,大多数知识图谱都具有不完整性,因此,知识图谱补全成为一项重要的任务。根据模型构造方法的不同,将知识图谱补全模型分为传统知识图谱补全模型、基于神经网络的知识图谱补全模型和基于元学习的知识图谱补全模型三类,对这三种知识图谱补全模型的分类情况进行介绍;总结知识图谱补全方法所使用的数据集和评价指标,并从各个模型优点和不足等方面对各类模型进行详细的对比分析。最后,对知识图谱补全进行归纳与总结,并展望未来的研究方向。

基于低秩分解和向量量化的深度网络压缩方法

随着人工智能的发展,深度神经网络成为多种模式识别任务中必不可少的工具,由于深度卷积神经网络(CNN)参数量巨大、计算复杂度高,将它部署到计算资源和存储空间受限的边缘计算设备上成为一项挑战。因此,深度网络压缩成为近年来的研究热点。低秩分解与向量量化是深度网络压缩中重要的两个研究分支,其核心思想都是通过找到原网络结构的一种紧凑型表达,从而降低网络参数的冗余程度。通过建立联合压缩框架,提出一种基于低秩分解和向量量化的深度网络压缩方法——可量化的张量分解(QTD)。该方法能够在网络低秩结构的基础上实现进一步的量化,从而得到更大的压缩比。在CIFAR-10数据集上对经典ResNet和该方法进行验证的实验结果表明,QTD能够在准确率仅损失1.71个百分点的情况下,将网络参数量压缩至原来的1%。而在大型数据集ImageNet上把所提方法与基于量化的方法PQF(Permute,Quantize,and Fine-tune)、基于低秩分解的方法TDNR(Tucker Decomposition with Nonlinear Response)和基于剪枝的方法CLIP-Q(Compression Learning by In-parallel Pruning-Quantization)进行比较与分析的实验结果表明,QTD能够在相同压缩范围下实现更好的分类准确率。