The Role and Place of Artificial Neural Network Architectures Structural Redundancy in the Input Data Prototypes and Generalization Development

Neural Networks (NN) are the functional unit of Deep Learning and are known to mimic the behavior of the human brain to solve complex data-driven problems. Whenever we train our own neural networks, we need to take care of something called the generalization of the neural network. The performance of Artificial Neural Networks (ANN) mostly depends upon its generalization capability. In this paper, we propose an innovative approach to enhance the generalization capability of artificial neural networks (ANN) using structural redundancy. A novel perspective on handling input data prototypes and their impact on the development of generalization, which could improve to ANN architectures accuracy and reliability is described.

An Intrusion Alarming System Based on Self-Similarity of Network Traffic

Intrusion detection system ean make effective alarm for illegality of networkusers, which is absolutely necessarily and important to build security environment of communicationbase service According to the principle that the number of network traffic can affect the degree ofself-similar traffic, the paper investigates the variety of self-similarity resulted fromunconventional network traffic. A network traffic model based on normal behaviors of user isproposed and the Hursl parameter of this model can be calculated. By comparing the Hurst parameterof normal traffic and the self-similar parameter, we ean judge whether the network is normal or notand alarm in time.

Load Reduction Test Method of Similarity Theory and BP Neural Networks of Large Cranes

Static load tests are an important means of supervising and detecting a crane＇s lift capacity. Due to space restrictions, however, there are difficulties and potential danger when testing large bridge cranes. To solve the loading problems of large-tonnage cranes during testing, an equivalency test is proposed based on the similarity theory and BP neural networks. The maximum stress and displacement of a large bridge crane is tested in small loads, combined with the training neural network of a similar structure crane through stress and displacement data which is collected by a physics simulation progressively loaded to a static load test load within the material scope of work. The maximum stress and displacement of a crane under a static load test load can be predicted through the relationship of stress, displacement, and load. By measuring the stress and displacement of small tonnage weights, the stress and displacement of large loads can be predicted, such as the maximum load capacity, which is 1.25 times the rated capacity. Experimental study shows that the load reduction test method can reflect the lift capacity of large bridge cranes. The load shedding predictive analysis for Sanxia 1200 t bridge crane test data indicates that when the load is 1.25 times the rated lifting capacity, the predicted displacement and actual displacement error is zero. The method solves the problem that lifting capacities are difficult to obtain and testing accidents are easily possible when 1.25 times related weight loads are tested for large tonnage cranes.

Mining Social Groups with Weighted Similarity in Campus Wireless Network

With the popularity of wireless networks and the prevalence of personal mobile computing devices, understanding the characteristic of wireless network users is of great significance to the network performance. In this study, system logs from two universities, Dartmouth College and Shanghai Jiao Tong University(SJTU), were mined and analyzed. Every user's log was represented by a user profile. A novel weighted social similarity was proposed to quantify the resemblance of users considering influence of location visits. Based on the similarity, an unsupervised learning method was applied to cluster users. Though environment parameters are different, two universities both form many social groups with Pareto distribution of similarity and exponential distribution of group sizes. These findings are very important to the research of wireless network and social network .

INTS-MFS:A novel method to predict microRNA-disease associations by integrating network topology similarity and microRNA function similarity

Identifying associations between microRNAs(miRNAs)and diseases is very important to understand the occurrence and development of human diseases.However,these existing methods suffer from the following limitation:first,some disease-related miRNAs are obtained from the miRNA functional similarity networks consisting of heterogeneous data sources,i.e.,disease similarity,protein interaction network,gene expression.Second,little approaches infer disease-related miRNAs depending on the network topological features without the functional similarity of miRNAs.In this paper,we develop a novel model of Integrating Network Topology Similarity and MicroRNA Function Similarity(INTS-MFS).The integrated miRNA similarities are calculated based on miRNA functional similarity and network topological characteristics.INTS-MFS obtained AUC of 0.872 based on five-fold cross-validation and was applied to three common human diseases in case studies.As a results,30 out of top 30 predicted Prostatic Neoplasm-related miRNAs were included in the two databases of dbDEMC and PhenomiR2.0.29 out of top 30 predicted Lung Neoplasm-related miRNAs and Breast Neoplasm-related miRNAs were included in dbDEMC,PhenomiR2.0 and experimental reports.Moreover,INTS-MFS found unknown association with hsa-mir-371a in breast cancer and lung cancer,which have not been reported.It provides biologists new clues for diagnosing breast and lung cancer.

A weight's agglomerative method for detecting communities in weighted networks based on weight's similarity

This paper proposes the new definition of the community structure of the weighted networks that groups of nodes in which the edge＇s weights distribute uniformly but at random between them. It can describe the steady connections between nodes or some similarity between nodes＇ functions effectively. In order to detect the community structure efficiently, a threshold coefficient t~ to evaluate the equivalence of edges＇ weights and a new weighted modularity based on the weight＇s similarity are proposed. Then, constructing the weighted matrix and using the agglomerative mechanism, it presents a weight＇s agglomerative method based on optimizing the modularity to detect communities. For a network with n nodes, the algorithm can detect the community structure in time O（n2 log~）. Simulations on networks show that the algorithm has higher accuracy and precision than the existing techniques. Furthermore, with the change of t~ the algorithm discovers a special hierarchical organization which can describe the various steady connections between nodes in groups.

Coarse Graining Method Based on Noded Similarity in Complex Network

Coarse graining of complex networks is an important method to study large-scale complex networks, and is also in the focus of network science today. This paper tries to develop a new coarse-graining method for complex networks, which is based on the node similarity index. From the information structure of the network node similarity, the coarse-grained network is extracted by defining the local similarity and the global similarity index of nodes. A large number of simulation experiments show that the proposed method can effectively reduce the size of the network, while maintaining some statistical properties of the original network to some extent. Moreover, the proposed method has low computational complexity and allows people to freely choose the size of the reduced networks.

基于相似网络和联合注意力的图嵌入模型

图注意力网络(graph attention network, GAT)将注意力机制与图神经网络融合,但模型只关注节点的一阶邻域节点,缺乏对高阶相似节点的考虑,同时在计算注意力分数时缺乏对节点结构特征的关注.为此提出一种基于相似网络和联合注意力的图嵌入模型.首先计算网络中的节点相似性,并将高相似度且未连接的节点对构建新边以形成相似网络.其次,引入结构相关性和内容相关性的概念,分别用于表征节点之间的结构关系和内容特征.通过融合两种相关性得分计算得到联合注意力分数.最后使用联合注意力分数对节点特征加权聚合,得到最终的节点嵌入表示.将本文所提算法在Cora、Citeseer和Pubmed 3个数据集上进行节点分类任务,准确率分别达到85.70%、74.30%、84.10%,与原始图注意力网络模型相比分别提高了2.70%、3.94%和2.60%.可见,所提出的算法可以得到更好的节点嵌入表示.

创新联合体潜在合作伙伴选择研究

为研究创新联合体构建过程中如何合理有效地挑选潜在合作伙伴,充分发挥创新联合体在实现高水平科技自立自强中的战略作用,构建了基于专利异构网络和SimRank算法的创新主体技术背景相似度计算方法,借助可视化方法展现潜在合作关系网络,结合谱聚类算法划分创新联合体潜在合作伙伴。在高分子材料领域进行实证分析,实证结果较好地展示了该领域创新主体的潜在合作关系和技术背景高度相似的创新主体聚类结果,证实了研究理论和方法的有效性,为创新联合体伙伴选择提供参考。

基于图结构增强的图神经网络方法

针对图卷积网络(GCNs)在面对低同质性的图结构时性能骤降问题,提出了一种新颖的基于图结构增强的图神经网络方法,用于学习改善的图节点表示。首先将节点信息通过消息传播和聚合,得到节点的初始表示;然后计算节点表示的相似性度量,得到图的同质结构;最后融合图的原始结构和同质结构进行节点的信息传递得到节点表示用于下游任务。结果表明:在6个公开的数据集上,所提算法在节点分类的多个指标上均优于对比算法,特别是在同质性较低的4个数据集上,所提算法的准确度(ACC)分数分别超过最高基准5.53%、6.87%、3.08%、4.00%,宏平均(F1)值分别超过最高基准5.75%、8.06%、6.46%、5.61%,获得了远高于基准的优越表现,表明所提方法成功改善了图数据的结构,验证了该算法对图结构优化的有效性。

基于动态异构网络的股价预测

股票预测通常被形式化为非线性的时间序列预测任务,但很少有研究者试图通过技术面数据去系统地揭示股票市场内在结构,例如股票上涨或下跌背后的原因可能是业务领域之间的合作或冲突,这些额外信息的增加有助于判断股票的未来趋势。为了充分真实刻画股票市场的交易状态,表达股票之间显式或隐式的关系,提出一种基于动态异构网络的股价预测模型sDHN(stock dynamic heterogeneous network),综合股票以及所属行业和地域,将其建模为动态异构网络。该模型在网络上引入动态时序特征,创新融合股票节点的四种不同技术层面的相似性图,生成富信息异构图,最后聚合不同元路径中隐含的语义信息生成嵌入,从异构图的角度充分探索股票之间的潜在关联。此外,在三个真实世界的股票数据集上进行了大量实验,所提出的模型准确率比所有基线模型均高出5%~34%,F_(1)-score则高出11.5%~37%,并且在图解释上证明了该方法的有效性。

基于PSO-BP神经网络的磨机传动系统模型修正

针对磨机传动系统结构的复杂性、部件间约束条件的不确定性以及非线性等因素,提出了一种基于PSO-BP神经网络的有限元模型修正方法。通过改进BP神经网络逼近设计参数和特征量间的非线性映射关系,结合实际结构响应,利用神经网络的泛化特性,得到了模型设计参数值。修正后频率误差从最高18%降到4%左右,修正系数误差范围均在0.5%以内,明显提高了有限元模型精度;同时,又不需要大量迭代求解步骤,避开了传统反问题模型修正法的复杂非线性优化过程,提升了效率,验证了PSO-BP神经网络法应用于大型磨机传动系统上的可行性,为后续传动系统整体分析奠定了基础。

基于图卷积的无监督跨模态哈希检索算法

为解决当前无监督跨模态哈希检索在全局相似性矩阵构建和异构数据语义信息融合中存在的困难,提出一种基于图卷积的无监督跨模态哈希检索算法(GCUH)。采用分层次聚合的方式,将各个模态的相似性结构编码到全局相似性矩阵中,获得跨模态的成对相似性信息来指导学习。使用图卷积模块融合跨模态信息,消除邻居结构中的噪声干扰,形成完备的跨模态表征,提出两种相似性保持的损失函数约束哈希码的一致性。与基线模型相比,GCUH在NUS-WIDE数据集上使用64位哈希码执行文本检索图片任务的检索精度提升了6.3%。

动量余弦相似度梯度优化图卷积神经网络

传统梯度下降算法仅对历史梯度进行指数加权累加,没有利用梯度的局部变化,造成优化过程越过全局最优解,即使收敛到最优解也会在最优解附近震荡,其训练图卷积神经网络会造成收敛速度慢、测试准确度低。利用相邻两次梯度的余弦相似度,动态调整学习率,提出余弦相似度梯度下降(SimGrad)算法。为进一步提升图卷积神经网络训练的收敛速度和测试准确度,减少震荡,结合动量思想提出动量余弦相似度梯度下降(NSimGrad)算法。通过收敛性分析,证明SimGrad算法、NSimGrad算法都具有O(√T)的遗憾界。在构建的三个非凸函数进行测试,并结合图卷积神经网络在四个数据集上进行实验,结果表明SimGrad算法保证了图卷积神经网络的收敛性,NSimGrad算法进一步提高图卷积神经网络训练的收敛速度和测试准确度,SimGrad、NSimGrad算法相较于Adam、Nadam具有更好的全局收敛性和优化能力。

基于双节点-双边图神经网络的茶叶病害分类方法

传统茶叶病害分类主要依赖人工方法,此类方法费工费时,同时茶叶病害样本较少使得现有的机器学习方法的模型训练不充分,病害分类准确率不够高。针对茶炭疽病、茶黑煤病、茶饼病和茶白星病4类病害,提出一种基于双节点-双边图神经网络的茶叶病害分类方法。首先通过两分支卷积神经网络提取RGB茶叶病害特征和灰度茶叶病害特征,两分支均采用ResNet12作为骨干网络,参数独立不共享,两类特征作为图神经网络的两个子节点,以获得不同域样本所包含的病害信息;其次构建相对度量边和相似性边两类边,从而强化节点对相邻节点所含病害特征的聚合能力。最后,经过双节点特征和双边特征更新模块,实现双节点和双边交替更新,提高边特征对节点距离度量的准确性,从而实现训练样本较少条件下对茶叶病害的准确分类。本文方法和小样本学习方法进行了对比实验,结果表明,本文方法获得更高的准确率,在miniImageNet和PlantVillage数据集上5way-1shot的准确率分别达到69.30%和88.42%,5way-5shot准确率分别为82.48%和93.04%。同时在茶叶数据集TeaD-5上5way-1shot和5way-5shot准确率分别达到84.74%和86.34%。

Functional Brain Network Learning Based on Spatial Similarity for Brain Disorders Identification

Functional brain network (FBN) measures based on functional magnetic resonance imaging (fMRI) data, has become important biomarkers for early diagnosis and prediction of clinical outcomes in neurological diseases, such as Alzheimer’s diseases (AD) and its prodromal state (i.e., Mild cognitive impairment, MCI). In the past decades, researchers have developed numbers of approaches for FBN estimation, including Pearson’s correction (PC), sparse representation (SR), and so on. Despite their popularity and wide applications in current studies, most of the approaches for FBN estimation only consider the dependency between the measured blood oxygen level dependent (BOLD) time series, but ignore the spatial relationships between pairs of brain regions. In practice, the strength of functional connection between brain regions will decrease as their distance increases. Inspired by this, we proposed a new approach for FBN estimation based on the assumption that the closer brain regions tend to share stronger relationships or similarities. To verify the effectiveness of the proposed method, we conduct experiments on a public dataset to identify the patients with MCIs from health controls (HCs) using the estimated FBNs. Experimental results demonstrate that the proposed approach yields statistically significant improvement in seven performance metrics over using the baseline methods.

基于特征相似性和特征规范化的注意力模块

近年来,注意力机制在图像分类、目标检测和语义分割等领域取得了巨大成功,但现有的注意力机制大多只能在通道或空间维度上实现特征融合,这极大限制了其在通道和空间维度上变化的灵活性,导致无法充分利用特征信息。为此,文中提出一种基于特征相似性和特征规范化的、可同时利用特征图各维度信息的卷积神经网络注意力模块FSNAM。该模块由特征相似性模块(FSM)和特征规范化模块(FNM)两部分组成,FSM利用输入特征图的通道特征信息和局部空间特征信息生成一个二维的特征相似性权重图;FNM利用输入特征图的全局空间特征信息生成一个三维的特征规范化权重图;两个模块生成的权重图融合在一起,生成一个三维的注意力权重图,以此实现通道特征信息和空间特征信息的融合。为证明FSNAM的可行性和有效性,进行了消融实验,结果表明:在图像分类任务方面,FSNAM模块对分类网络在CIFAR数据集上的性能提升明显优于其他主流注意力模块;在目标检测任务方面,使用FSNAM模块的目标检测网络对VOC数据集中的小目标和中等大小目标的检测准确率分别提高了3.9和1.2个百分点;在语义分割任务方面,使用FSNAM模块可以提高HRNet模型的性能,在SBD数据集上模型的平均像素准确率提高了0.58个百分点。

基于节点相似性的二阶链路预测方法

复杂网络中基于节点相似性的链路预测算法通常根据两个节点之间的相似度,预测节点对之间是否存在链路。提出基于节点相似性的二阶链路预测方法,判别节点对之间是否存在未连接的节点,并补全节点对之间的二阶链路。同时,提出二阶链路预测指标,计算已知节点与其他并不存在链路的节点之间的相似性,并构建二阶可达网络保留原始网络中的二阶链路信息。实验结果表明,该方法能够在真实的网络数据中找到节点对之间的缺失节点,并补全可能存在的二阶链路。不同的链路预测指标在4个不同网络中的性能表现有所不同,所有实验中的最佳精确率达83.7%。

相似网络构建与表征的水下声信号检测

水下声信号检测在海洋防御系统中扮演着不可或缺的角色,同时也广泛应用于民用领域。然而,在没有目标信号先验信息的情况下,目前仍缺乏行之有效的水下声信号检测方法。为此,该文提出了一种新的算法—相似网络,以解决在复杂海洋背景下水下目标检测的难题。该方法结合了信息几何和复杂网络理论,通过将节点相似度度量问题转化为矩阵流形上的几何问题,测量不同时间尺度上数据之间的相似性,并构建时间序列数据的网络表示。同时还引入了图信号处理理论,以提取目标信号内部隐藏的动力学特性,从而实现无目标先验信息下的水下声信号检测。通过对仿真和实测数据的研究验证,证明了该方法的有效性。结果表明,相似网络方法优于现有的网络构建和目标信号被动检测方法,能够更有效地检测水下声信号,实现无目标先验信息下的水下声信号检测。