Application of graph neural network and feature information enhancement in relation inference of sparse knowledge graph

At present,knowledge embedding methods are widely used in the field of knowledge graph(KG)reasoning,and have been successfully applied to those with large entities and relationships.However,in research and production environments,there are a large number of KGs with a small number of entities and relations,which are called sparse KGs.Limited by the performance of knowledge extraction methods or some other reasons(some common-sense information does not appear in the natural corpus),the relation between entities is often incomplete.To solve this problem,a method of the graph neural network and information enhancement is proposed.The improved method increases the mean reciprocal rank(MRR)and Hit@3 by 1.6%and 1.7%,respectively,when the sparsity of the FB15K-237 dataset is 10%.When the sparsity is 50%,the evaluation indexes MRR and Hit@10 are increased by 0.8%and 1.8%,respectively.

ISAR target recognition based on non-negative sparse coding

Aiming at technical difficulties in feature extraction for the inverse synthetic aperture radar （ISAR） target recognition, this paper imports the concept of visual perception and presents a novel method, which is based on the combination of non-negative sparse coding （NNSC） and linear discrimination optimization, to recognize targets in ISAR images. This method implements NNSC on the matrix constituted by the intensities of pixels in ISAR images for training, to obtain non-negative sparse bases which characterize sparse distribution of strong scattering centers. Then this paper chooses sparse bases via optimization criteria and calculates the corresponding non-negative sparse codes of both training and test images as the feature vectors, which are input into k neighbors classifier to realize recognition finally. The feasibility and robustness of the proposed method are proved by comparing with the template matching, principle component analysis （PCA） and non-negative matrix factorization （NMF） via simulations.

Graph Regularized L_p Smooth Non-negative Matrix Factorization for Data Representation

This paper proposes a Graph regularized Lpsmooth non-negative matrix factorization(GSNMF) method by incorporating graph regularization and L_p smoothing constraint, which considers the intrinsic geometric information of a data set and produces smooth and stable solutions. The main contributions are as follows: first, graph regularization is added into NMF to discover the hidden semantics and simultaneously respect the intrinsic geometric structure information of a data set. Second,the Lpsmoothing constraint is incorporated into NMF to combine the merits of isotropic(L_2-norm) and anisotropic(L_1-norm)diffusion smoothing, and produces a smooth and more accurate solution to the optimization problem. Finally, the update rules and proof of convergence of GSNMF are given. Experiments on several data sets show that the proposed method outperforms related state-of-the-art methods.

Jointly-check iterative decoding algorithm for quantum sparse graph codes

For quantum sparse graph codes with stabilizer formalism, the unavoidable girth-four cycles in their Tanner graphs greatly degrade the iterative decoding performance with standard belief-propagation （BP） algorithm. In this paper, we present a jointly-check iterative algorithm suitable for decoding quantum sparse graph codes efficiently. Numerical simulations show that this modified method outperforms standard BP algorithm with an obvious performance improvement.

Graph Regularized Sparse Coding Method for Highly Undersampled MRI Reconstruction

The imaging speed is a bottleneck for magnetic resonance imaging( MRI) since it appears. To alleviate this difficulty,a novel graph regularized sparse coding method for highly undersampled MRI reconstruction( GSCMRI) was proposed. The graph regularized sparse coding showed the potential in maintaining the geometrical information of the data. In this study, it was incorporated with two-level Bregman iterative procedure that updated the data term in outer-level and learned dictionary in innerlevel. Moreover,the graph regularized sparse coding and simple dictionary updating stages derived by the inner minimization made the proposed algorithm converge in few iterations, meanwhile achieving superior reconstruction performance. Extensive experimental results have demonstrated GSCMRI can consistently recover both real-valued MR images and complex-valued MR data efficiently,and outperform the current state-of-the-art approaches in terms of higher PSNR and lower HFEN values.

Modeling of unsupervised knowledge graph of events based on mutual information among neighbor domains and sparse representation

Text event mining,as an indispensable method of text mining processing,has attracted the extensive attention of researchers.A modeling method for knowledge graph of events based on mutual information among neighbor domains and sparse representation is proposed in this paper,i.e.UKGE-MS.Specifically,UKGE-MS can improve the existing text mining technology's ability of understanding and discovering high-dimensional unmarked information,and solves the problems of traditional unsupervised feature selection methods,which only focus on selecting features from a global perspective and ignoring the impact of local connection of samples.Firstly,considering the influence of local information of samples in feature correlation evaluation,a feature clustering algorithm based on average neighborhood mutual information is proposed,and the feature clusters with certain event correlation are obtained;Secondly,an unsupervised feature selection method based on the high-order correlation of multi-dimensional statistical data is designed by combining the dimension reduction advantage of local linear embedding algorithm and the feature selection ability of sparse representation,so as to enhance the generalization ability of the selected feature items.Finally,the events knowledge graph is constructed by means of sparse representation and l1 norm.Extensive experiments are carried out on five real datasets and synthetic datasets,and the UKGE-MS are compared with five corresponding algorithms.The experimental results show that UKGE-MS is better than the traditional method in event clustering and feature selection,and has some advantages over other methods in text event recognition and discovery.

Two-Level Bregman Method for MRI Reconstruction with Graph Regularized Sparse Coding

In this paper, a two-level Bregman method is presented with graph regularized sparse coding for highly undersampled magnetic resonance image reconstruction. The graph regularized sparse coding is incorporated with the two-level Bregman iterative procedure which enforces the sampled data constraints in the outer level and updates dictionary and sparse representation in the inner level. Graph regularized sparse coding and simple dictionary updating applied in the inner minimization make the proposed algorithm converge with a relatively small number of iterations. Experimental results demonstrate that the proposed algorithm can consistently reconstruct both simulated MR images and real MR data efficiently, and outperforms the current state-of-the-art approaches in terms of visual comparisons and quantitative measures.

稀疏图的r-动态染色

通过分析极小反例的结构性质,运用权转移的方法,研究了对于mad(G)<14/5的稀疏图G的r-动态染色数,证明了对于满足mad(G)<14/5的图G,若r≥9,则χr(G)≤r+2.研究结果推广了稀疏图r-动态染色的已知结果.

多任务联合学习的图卷积神经网络推荐

基于图神经网络的协同过滤推荐可以更有效地挖掘用户项目之间的交互信息,但其性能依然受到数据稀疏和表征学习质量不高问题的影响,因此提出一种多任务联合学习的图卷积神经网络推荐(multi-task joint learning for graph convolutional neural network recommendations,MTJL-GCN)模型。利用图神经网络在用户-项目交互图上所聚集到的同质结构信息与初始嵌入信息形成结构邻居关系,设计节点邻居关系的对比学习辅助任务来缓解数据稀疏问题;向节点的原始表征添加随机的统一噪声进行表征级数据增强,构建节点表征关系的对比学习辅助任务,并提出直接优化对齐性和均匀性两个属性的学习目标来提高表征学习质量;将图协同过滤推荐任务与对比学习辅助任务和直接优化学习目标进行联合训练,从而提升推荐性能。在Amazon-books和Yelp2018两个公开数据集上进行实验,该模型在Recall@k和NDCG@k两个推荐性能指标上的表现均优于基线模型,证明了MTJL-GCN模型的有效性。

基于Distance-2算法的并行Jacobian矩阵计算及其在耦合问题中的应用

并行Newton-Krylov方法是求解大规模多物理耦合问题的有效方法,如何高效自动计算Jacobian矩阵是一大难点。利用有限差分方法,可避免推导Jacobian矩阵的表达式,实现矩阵的自动计算。现有工作表明,在串行环境下利用矩阵的稀疏性和图着色算法,Jacobian矩阵的计算效率可提高至少1个量级。但在并行环境下,串行着色算法失效,需采用相应的并行着色算法。本研究将图论领域的Distance-2算法应用于Jacobian矩阵的并行着色。通过求解一个简化多物理耦合问题检验了该并行算法的正确性和计算效率。测试结果表明,该并行算法得到的Jacobian矩阵完全正确;着色数随着并行核数的增加略微有所增加,100个进程下并行效率为56%;基于该算法求解多物理耦合问题,其计算时间和Krylov迭代次数较JFNK减少了约1/2。

稀疏分解和图拉普拉斯正则化的图像前景背景分割方法

针对现有图像前景背景分割方法的分割结果存在孤立像素点的问题,利用图信号处理理论和稀疏分解模型,提出新的图像前景背景分割方法.将图像的内在结构建模为图,通过图模型有效地刻画像素之间的内在关联性.将图像的像素强度建模为图信号,其中图像背景作为平滑分量,由一组图傅里叶变换基函数线性表示,叠加在背景上的前景为稀疏分量,前景像素间的连通性可由图拉普拉斯正则化项进行刻画.将图像前景背景分割问题归结为包含稀疏分解模型和图拉普拉斯正则化项的约束优化问题,采用交替方向乘子法对该优化问题进行求解.实验结果表明,与现有的其他方法相比,所提方法具有更好的分割效果.

基于图滤波与自表示的无监督特征选择算法

针对现有方法未考虑数据的高阶邻域信息而不能完全捕捉数据内在结构的问题,提出一种基于图滤波与自表示的无监督特征选择算法.首先,将高阶图滤波器应用于数据获得其平滑表示,并设计一个正则化器联合高阶图信息进行自表示矩阵学习以捕捉数据的内在结构;其次,应用l_(2,1)范数重建误差项和特征选择矩阵,以增强模型的鲁棒性与稀疏性选择判别的特征;最后,用一个迭代算法有效地求解所提出的目标函数,并进行仿真实验以验证该算法的有效性.

OFDM叠加导频联合信道估计和检测方法

针对现有正交频分复用(Orthogonal Frequency Division Multiplexing,OFDM)系统信道估计和迭代检测算法中频谱效率低和鲁棒性差等问题,提出了一种基于酉近似消息传递和叠加导频的信道估计与联合检测方法。首先,在软调制/解调中叠加导频对正交幅度调制的星座点进行预处理,检测时将叠加的导频作为频域符号的先验分布,利用置信传播算法进行调制和解调,实现检测模型的简化。然后,应用因子图-消息传递算法对OFDM传输系统和信道进行建模和全局优化,引入酉变换加强信道估计算法的鲁棒性。最后,建立OFDM仿真环境对现有方法进行仿真分析。仿真结果表明,相对于现有的独立导频类算法,所提算法能够以相同复杂度显著提升OFDM系统的频谱效率和鲁棒性。

Improved Non-negative Matrix Factorization Algorithm for Sparse Graph Regularization

Aiming at the low recognition accuracy of non-negative matrix factorization(NMF)in practical application,an improved spare graph NMF(New-SGNMF)is proposed in this paper.New-SGNMF makes full use of the inherent geometric structure of image data to optimize the basis matrix in two steps.A threshold value s was first set to judge the threshold value of the decomposed base matrix to filter the redundant information in the data.Using L2 norm,sparse constraints were then implemented on the basis matrix,and integrated into the objective function to obtain the objective function of New-SGNMF.In addition,the derivation process of the algorithm and the convergence analysis of the algorithm were given.The experimental results on COIL20,PIE-pose09 and YaleB database show that compared with K-means,PCA,NMF and other algorithms,the proposed algorithm has higher accuracy and normalized mutual information.

结合最近邻图模型的稀疏ISAR成像方法

逆合成孔径雷达(Inverse Synthetic Aperture Radar,ISAR)稀疏成像方法可提供图像对比度高、旁瓣干扰少的成像结果 .稀疏成像以场景或目标散射率分布具有稀疏性为前提,待成像目标场景的稀疏特性决定了最终成像质量. ISAR目标场景的自然稀疏特性着重刻画点状特征,变换域稀疏表示可增强目标图像的纹理等通用特征.通过学习获得的稀疏变换字典,可自适应于待成像的ISAR目标场景,找到面向ISAR目标图像块的特有稀疏表示.但是,图像块的特有稀疏表示中忽略了待成像目标场景中目标的几何特征信息.最近邻图模型可建立给定数据的几何特征描述算子,刻画出给定数据的几何特征信息.本文利用最近邻图模型来刻画待成像目标场景中目标的几何特征信息,并映射到待成像目标场景的特有稀疏表示中;提出结合最近邻图模型的ISAR稀疏成像方法,用于不同类别实测ISAR数据成像.相比已有的ISAR稀疏成像方法,所提成像方法可获得目标轮廓更清晰的成像结果,成像所需时间平均减少10.4%.

基于L1-Graph表示的标记传播多观测样本分类算法

同类样本被认为是分布在同一个高维观测空间的低维流形上,针对多观测样本分类如何利用这一流形结构的问题,提出基于L1-Graph表示的标记传播多观测样本分类算法。首先基于稀疏表示的思路构造L1-Graph,进而得到样本之间的相似度矩阵,然后在半监督分类标记传播算法的基础上,限制所有的观测样本都属于同一个类别的条件下,得到一个具有特殊结构的类标矩阵,最后把寻找最优类标矩阵的计算转化为离散目标函数优化问题,进而计算出测试样本所属类别。在USPS手写体数据库、ETH-80物体识别数据库以及Cropped Yale人脸识别数据库上进行了一系列实验,实验结果表明了本文提出方法的可行性和有效性。

域自适应动态图卷积网络下的地铁客流预测

针对客流预测中存在因数据量有限导致模型训练过程中出现高方差和泛化性差等问题,提出一种域自适应动态图卷积网络(GCN-DANN)。通过构建地铁线路的节点网络拓扑结构,并利用动态图卷积网络提取相邻站点之间的流量、站点所属线路的交通负载以及不同线路之间的流量传播等关联特征。同时采用迁移学习自适应对齐源域和目标域的特征,减少因数据分布不一致而导致预测性能低等现象。最后,通过全连接层将源域和目标域中的特征进行信息融合,进而弥补训练过程出现高方差和泛化性差等缺陷。在深圳地铁数据集上对模型训练,分别在杭州地铁全样本和20%样本数据集上进行测试和验证。实验结果表明,在20%样本数据集下,GCN-DANN网络与经典预测网络相比,MAE、RMSER和MAPE分别平均下降5.34%、6.07%、2.97%。在全样本数据集下,GCN-DANN在20%样本基础上的三项指标分别下降2.76%、1.77%、3.5%,相较于其他经典网络下降幅度最小。研究可解决实际应用中因数据稀缺导致预测效果差的问题。

Encoding of rat working memory by power of multi-channel local field potentials via sparse non-negative matrix factorization

Working memory plays an important role in human cognition. This study investigated how working memory was encoded by the power of multichannel local field potentials （LFPs） based on sparse non negative matrix factorization （SNMF）. SNMF was used to extract features from LFPs recorded from the prefrontal cortex of four SpragueDawley rats during a memory task in a Y maze, with 10 trials for each rat. Then the powerincreased LFP components were selected as working memoryrelated features and the other components were removed. After that, the inverse operation of SNMF was used to study the encoding of working memory in the time frequency domain. We demonstrated that theta and gamma power increased significantly during the working memory task. The results suggested that postsynaptic activity was simulated well by the sparse activity model. The theta and gamma bands were meaningful for encoding working memory.

Graph-Based Dimensionality Reduction for Hyperspectral Imagery: A Review

Hyperspectral image(HSI)contains a wealth of spectral information,which makes fine classification of ground objects possible.In the meanwhile,overly redundant information in HSI brings many challenges.Specifically,the lack of training samples and the high computational cost are the inevitable obstacles in the design of classifier.In order to solve these problems,dimensionality reduction is usually adopted.Recently,graph-based dimensionality reduction has become a hot topic.In this paper,the graph-based methods for HSI dimensionality reduction are summarized from the following aspects.1)The traditional graph-based methods employ Euclidean distance to explore the local information of samples in spectral feature space.2)The dimensionality-reduction methods based on sparse or collaborative representation regard the sparse or collaborative coefficients as graph weights to effectively reduce reconstruction errors and represent most important information of HSI in the dictionary.3)Improved methods based on sparse or collaborative graph have made great progress by considering global low-rank information,local intra-class information and spatial information.In order to compare typical techniques,three real HSI datasets were used to carry out relevant experiments,and then the experimental results were analysed and discussed.Finally,the future development of this research field is prospected.

融合图神经网络和稀疏自注意力的会话推荐分析

针对现有会话推荐采用单一模型无法兼顾全局和局部信息,从而影响推荐性能的问题,提出融合图神经网络和稀疏自注意力的会话推荐模型(SSA-GNN)。模型采用稀疏自注意力构建全局隐向量,以解决无关项的干扰和图神经网络难以表示长距离依赖的问题;采用目标注意图神经网络构建局部隐向量,更深层次的捕获项目间的复杂依赖。最后在预测层将全局和局部隐向量线性连接,有效兼顾了全局和局部信息。模型在Yoochoose1/64数据集上的试验结果比基线模型GC-SAN在评价指标P@20上提高了1.25%,MRR@20上提高了4.59%。