基于改进注意力机制的实体关系抽取方法

实体关系抽取是知识库构建中至关重要的一个环节.在众多的实体关系抽取方法中,远程监督结合神经网络模型的方法在准确率等性能上是比较令人满意的,但远程监督获取的标注语料中往往存在大量的噪声数据,给实体关系抽取模型的训练带来了很大的影响.本文提出一种基于改进注意力机制的卷积神经网络实体关系抽取模型.该模型针对包含同一实体对的句子集合,从中尽可能地找出所有体现该实体对关系的正实例,构建组合句子向量,抛弃可能的噪声句子,从而最大程度地降低噪声句子的影响又能充分利用正实例的语义信息.实验证明,本文提出的关系抽取模型在准确率上优于对比的关系抽取模型.     

Knowledge triple extraction in cybersecurity with adversarial active learning

Aiming at the problem that using pipeline methods for extracting cybersecurity knowledge triples may cause the errors propagation of entity recognition and did not consider the correlation between entity recognition and relation extraction,and training triple extraction model lacked labeled corpora,an end-to-end cybersecurity knowledge triple extraction method with adversarial active learning was proposed.For knowledge triple extraction,the conventional entity recognition and relation extraction were modelled as sequence labeling task through joint labeling strategy firstly.And then,a BiLSTM-LSTM-based model with dynamic attention mechanism was designed to jointly extract entities and relations,forming triples.Finally,with adversarial learning framework,a discriminator was trained to incrementally select high-quality samples for labeling,and the performance of the joint extraction model was continuously enhanced by iterative retraining.Experiments show that the proposed joint extraction model outperforms the existing cybersecurity knowledge triple extraction methods,and demonstrate the effectiveness of proposed adversarial active learning scheme.     

深度学习下的散射成像:物理与数据联合建模优化（特邀）

为了利用被散射的光信号实现成像，越来越多的散射成像方法被提出。其中深度学习以其强大的数据表征和信息提取能力在散射成像领域发挥着重要的作用。相较于传统散射成像方法，基于深度学习的散射成像方法在成像速度、质量、信息维度等方面都有着巨大的优势。但是，模型训练、模型泛化等问题也制约着该方法的发展。因此，越来越多的研究将物理过程与基于数据驱动的方法进行联合建模，利用物理先验指导神经网络优化。相较于单纯的数据驱动方法而言，物理-数据联合建模的方法对数据量、神经网络参数量的依赖程度大大降低，在保证成像质量的前提下有效降低数据获取难度及对实验环境的要求。联合建模优化的方式实现了介质、目标类型等散射成像中关键节点的泛化。同时在训练过程方面，实现了从有监督到半监督再到无监督的训练优化过程迭代，不同模型和监督方式的提出大大提升了基于深度学习方法的训练效率，在降低对硬件和时间成本的同时，提升了基于深度学习的散射成像方法在非实验室场景应用的可能性。     

面向微博的多实体稀疏关系数据联合聚类

针对大规模微博中多实体间的稀疏关系数据,提出一种面向多实体稀疏关系数据的高效联合聚类算法。在算法中,为了充分利用多关系数据,提出了一种顽健的约束信息嵌入方法构建关系矩阵,降低了矩阵的稀疏性,进一步提高了算法的准确率。在稀疏约束的块坐标下降框架下,关系矩阵通过非负矩阵三分解算法同时获得不同实体的聚类指示矩阵。非负矩阵分解过程中,通过高效的投射算法实现快速求解,确保了聚类结果的稀疏结构。在人工和真实数据集上的实验表明,算法在3个指标上都具有明显提高,特别是在极端稀疏数据上的效果更加明显。     

Multiscale residual gradient attention for face anti-spoofing

In the field of face anti-spoofing (FAS), how to extract the representative features to distinguish between real and spoof faces and train the corresponding deep networks are two vital issues. In this paper, we propose a simple but effective end-to-end FAS model based on an innovative texture extractor and a depth auxiliary supervision mechanism. In the feature extraction stage, we first design the residual gradient convolutions based on the redesigned gradient operators, which are used to extract fine-grained texture features. The extraction of texture features is based on multiple scales by dividing the texture differences between living and spoofing faces into three levels reasonably. Then we construct a multiscale residual gradient attention (MRGA) to obtain representative texture features from multiple levels texture features. By combining the proposed feature extractor MRGA and existing vision transformer (ViT), the MRGA-ViT is proposed to generate related semantics and obtain final classification results. In the training stage, we also propose a local depth auxiliary supervision based on a novel adjacent depth loss, which utilizes the correlation information of adjacent pixels adequately compared with traditional depth loss. The proposed MRGA-ViT model achieves competitive performance in generalization and stability ability, e.g., the ACER(%) values of intra testing on OULU-NPU database are 1.8, 2.6, 1.6 ± 1.2 and 1.9 ± 2.7 respectively, the AUC(%) of cross type testing attains 99.45 ± 0.57, the ACER(%) values of cross dataset testing are 28.1 and 36.7 respectively. Experimental results prove that the proposed model is competitive to other state-of-the-art works on generalization and stability performance.     

基于机器学习的端口扫描入侵检测

在入侵检测系统发展的30年间,不断有新的检测方法被提出。在如今的第四次工业革命——人工智能的潮流中,机器学习算法为各种系统的方法解决提供了新的思路。基于2018年Daniel Fraunholz等人提出了的入侵检测模型,提出了一种基于机器学习的端口扫描检测系统,其中系统的特征提取参考了KDD Cup 99数据集中数据的特征提取,而其中的模型训练集是基于CICIDS2017数据集的。最后,模型测试结果优良。     

Wind speed prediction based on nested shared weight long short-term memory network

With the expansion of wind speed data sets, decreasing model training time is of great significance to the time cost of wind speed prediction. And imperfection of the model evaluation system also affect the wind speed prediction. To address these challenges, a hybrid method based on feature extraction, nested shared weight long short-term memory(NSWLSTM) network and Gaussian process regression(GPR) was proposed. The feature extraction of wind speed promises the best performance of the model. NSWLSTM model reduces the training time of long short-term memory(LSTM) network and improves the prediction accuracy. Besides, it adopted a method combined NSWLSTM with GPR(NSWLSTM-GPR) to provide the probabilistic prediction of wind speed. The probabilistic prediction can provide information that deviates from the predicted value, which is conducive to risk assessment and optimal scheduling. The simulation results show that the proposed method can obtain high-precision point prediction, appropriate prediction interval and reliable probabilistic prediction results with shorter training time on the wind speed prediction.     

Skeleton-based deep pose feature learning for action quality assessment on figure skating videos

Most of the existing Action Quality Assessment (AQA) methods for scoring sports videos have deeply researched how to evaluate the single action or several sequential-defined actions that performed in short-term sport videos, such as diving, vault, etc. They attempted to extract features directly from RGB videos through 3D ConvNets, which makes the features mixed with ambiguous scene information. To investigate the effectiveness of deep pose feature learning on automatically evaluating the complicated activities in long-duration sports videos, such as figure skating and artistic gymnastic, we propose a skeleton-based deep pose feature learning method to address this problem. For pose feature extraction, a spatial–temporal pose extraction module (STPE) is built to capture the subtle changes of human body movements and obtain the detail representations for skeletal data in space and time dimensions. For temporal information representation, an inter-action temporal relation extraction module (ATRE) is implemented by recurrent neural network to model the dynamic temporal structure of skeletal subsequences. We evaluate the proposed method on figure skating activity of MIT-skate and FIS-V datasets. The experimental results show that the proposed method is more effective than RGB video-based deep feature learning methods, including SENet and C3D. Significant performance progress has been achieved for the Spearman Rank Correlation (SRC) on MIT-Skate dataset. On FIS-V dataset, for the Total Element Score (TES) and the Program Component Score (PCS), better SRC and MSE have been achieved between the predicted scores against the judge’s ones when compared with SENet and C3D feature methods.     

铁电薄膜在热 电 力耦合场下的本构模型

为了更好地描述铁电薄膜在外场下的非线性畴变行为,该文提出了铁电薄膜在热-电-力耦合场下的本构模型。该模型基于细观力学模型,认为电畴自由能的改变提供电畴翻转的动力。在铁电材料的本构关系中加入了铁电薄膜制备过程中产生的残余应变项,以便描述铁电薄膜材料在外加热 电 力耦合场下的非线性行为。模型的理论计算结果与实验结果相符,并与其他模型中的计算结果一致。     

基于深度学习的通信辐射源指纹特征提取算法

针对在小样本条件下难以有效提取通信辐射源指纹特征的问题，设计了一种堆栈自编码网络的通信辐射源个体细微特征提取算法。首先通过预处理（高阶谱分析）将原始通信辐射源信号从时域转化到高维特征空间，然后利用大量无标签的通信辐射源高维样本训练堆栈自编码器网络，在此基础上，通过少量有标签的通信辐射源样本对softmax回归模型进行精校训练，从而获得面向通信辐射源指纹特征提取的深度学习网络。实际采集的通信电台数据集上的实验结果验证了该模型的可行性与有效性。      

多尺度混合注意力胶囊网络的海洋鱼类识别

针对胶囊网络(capsule network,CapsNet)特 征提取结构单一和数据处理中参数量过大的问题,提出 多尺度混合注意力胶囊网络 模型。首先,在网络初始端添加不同尺度的卷积核来多角度提取 特征,并引 入混合注意力机制,通过聚焦更具分辨性的特征区域来降低复杂背景干扰。其次,采用局部 剪枝算法优 化动态路由,减少参数量,缩短模型训练时间。最后,在海洋鱼类数据集F4K(Fish4Knowled ge)上验证, 结果表明,与传统残差网络(residual network50,ResNet-50)、双线性网络(bilinear convo l utional neural network,B-CNN)、分层精简双线性注意力网络(spatial transformation netw ork and hierarchical compact bilinear pooling,STN-H-CBP)以及CapsNet模型相比,该算法 识别精度为98.65%,比ResNet-50模型提升 了5.92%;训练时间为2.2 h,相比于CapsNet 缩短了近40 min,验证了该算法的可行性。     

BP网络学习能力与泛化能力之间的定量关系式

分析BP网络过拟合时网络学习能力与泛化能力之间的内在联系,引入描述问题复杂性程度的复相关系数,建立了BP网络过拟合时,反映网络学习能力的训练样本集的训练相对误差与表征泛化能力的网络对检验样本集的测试相对误差之间满足的定量关系式.通过模拟若干不同类型函数的BP网络数值建模试验,确定了关系式中过拟合参数q的取值范围为0.007~0.07,指出BP网络应用于给定样本集的训练过程中,具有较佳泛化能力的停止训练方法.     

人际网络关系抽取和结构挖掘

社会是一个由多种多样的关系构成的巨大网络.对社会网络进行研究,可以揭示关系的结构,解释一定的社会现象.为此首先引入关系以及关系描述词的定义,利用关系抽取技术,扩展传统的二元关系,提出了一种基于同义词词林的提取关系描述词的方法,并收集特定领域内的人物关系信息构建成人际网络.接着对人际关系网络的拓扑结构,社区发现以及网络社区核心人物进行了研究,同时对分析结果实现了可视化.结果表明,社会网络分析可以揭示真实社会中的许多现象,有助于人们理解和开发这些网络.     

机载LiDAR点云中电力线的提取和重建研究

为了实现输电线走廊的有效监管，采用了一种基于机载激光雷达(LiDAR)点云的电力线提取和重建的方法。首先利用点云的回波信息滤除大部分地物点，保留电力线的全部信息, 然后基于格网高差和高程阈值剔除大部分非电力线点云，实现电力线的初提取。针对初步提取出的电力线点云，通过Hough变换分离出单根输电线点。为提高电力线点的提取精度，对Hough变换提取的结果进行了改进，利用局部极值点检测的方法确定杆塔的位置，最后通过带有限制条件的多项式模型对每档电力线进行分段拟合，实现电力线模型的3维重建。结果表明，利用该算法提取出了424个电力线点，总的提取精度达到87.603%，拟合出的电力线模型效果较好。该算法可以自动、精确地实现LiDAR点云数据中电力线的提取和重建，对输电线走廊巡检具有很好的实用价值。     

High-quality voice conversion system based on GMM statistical parameters and RBF neural network

A voice conversion （VC） system was designed based on Gaussian mixture model （GMM） and radial basis function （RBF） neural network. As a voice conversion model, RBF network needs quantities of training data to improve its performance. For one speech, the networks trained by different segments of data have different transformation effects. Since trying segment by segment to obtain the best conversion effect is complex, a conversion method was proposed, that uses GMM for statistics before training RBF network to aim at the problem. The speech transformation and representation using adaptive interpolation of weighted spectrum （STRAIGHT） model is used for accurate extraction of vocal tract spectrum. Then GMM is used to classify the numerous spectral parameters. The obtained mean parameters were trained in RBF network. Experiment reveals that, the soft classification ability of GMM can promptly realize the reduction and classification of training data under the premise of ensuring the training effect. The selection complexity is decreased thereafter. Compared to the conventional RBF network training methods, this method can make the transformation of spectral parameters more effective and improve the quality of converted speech.     

KFDA and clustering based multiclass SVM for intrusion detection

To improve the classification accuracy and reduce the training time, an intrusion detection technology is proposed, which combines feature extraction technology and multiclass support vector machine （SVM） classification algorithm. The intrusion detection model setup has two phases. The first phase is to project the original training data into kernel fisher discriminant analysis （KFDA） space. The second phase is to use fuzzy clustering technology to cluster the projected data and construct the decision tree, based on the clustering results. The overall detection model is set up based on the decision tree. Results of the experiment using knowledge discovery and data mining （KDD） from 99 datasets demonstrate that the proposed technology can be an an effective way for intrusion detection.     

基于主观逻辑的群体动态信任模型

目前信任模型的研究仅仅涉及一对一的个体信任关系,并没有充分考虑对一对多、多对一以及多对多的群体信任关系的支持.从信任的主观性角度出发,提出了4种基本群体约束模式及其嵌套复合来刻画群体构成约束,在此基础上提出了一种基于主观逻辑的群体信任模型,给出了4种基本约束模式及其嵌套复合的直接信任度和推荐信任度的度量公式,对于群体信任关系,特别是软件群体的信任关系提供了很好的支持.通过模型的表达能力和有效性分析表明:群体信任模型具有较好的适用性和实用性.     

基于密集网络的质子热声信号走时提取

针对临床上由质子热声信号脉宽和信噪比的不确定性引起的走时提取困难问题,提出了一种基于密集网络的走时提取算法.该算法使用密集块代替传统卷积块,融合了具有不同感受野的特征,并引入了深度监督和网络剪枝机制,利用标记好的质子束热声信号数据进行学习,以提取所需的走时信息.实验结果表明,相比其他算法,该算法对质子热声信号走时的提取具有较高的准确率和鲁棒性,同时展现了实时提取的可行性.     

基于多模态特征融合监督的RGB-D图像显著性检测

RGB-D图像显著性检测是在一组成对的RGB和Depth图中识别出视觉上最显著突出的目标区域。已有的双流网络,同等对待多模态的RGB和Depth图像数据,在提取特征方面几乎一致。然而,低层的Depth特征存在较大噪声,不能很好地表征图像特征。因此,该文提出一种多模态特征融合监督的RGB-D图像显著性检测网络,通过两个独立流分别学习RGB和Depth数据,使用双流侧边监督模块分别获取网络各层基于RGB和Depth特征的显著图,然后采用多模态特征融合模块来融合后3层RGB和Depth高维信息生成高层显著预测结果。网络从第1层至第5层逐步生成RGB和Depth各模态特征,然后从第5层到第3层,利用高层指导低层的方式产生多模态融合特征,接着从第2层到第1层,利用第3层产生的融合特征去逐步地优化前两层的RGB特征,最终输出既包含RGB低层信息又融合RGB-D高层多模态信息的显著图。在3个公开数据集上的实验表明,该文所提网络因为使用了双流侧边监督模块和多模态特征融合模块,其性能优于目前主流的RGB-D显著性检测模型,具有较强的鲁棒性。