一种基于因子图模型的半监督社区发现方法

社区发现是社交网络分析中一个重要的研究方向.当前大部分的研究都聚焦在自动社区发现问题,但是在具有数据缺失或噪声的网络中,自动社区发现算法的性能会随着噪声数据的增加而迅速下降.通过在社区发现中融合先验信息,进行半监督的社区发现,有望为解决上述挑战提供一条可行的途径.本文基于因子图模型,通过融入先验信息到一个统一的概率框架中,提出了一种基于因子图模型的半监督社区发现方法,研究具有用户引导情况下的社交网络社区发现问题.在三个真实的社交网络数据（Zachary社会关系网、海豚社会网和DBLP协作网）上进行实验,证明通过融入先验信息可以有效地提高社区发现的精度,且将我们的方法与一种最新的半监督社区发现方法（半监督Spin-Glass模型）进行对比,在三个数据集中F-measure平均提升了6.34%、16.36%和12.13%.     

一种基于广义期望首达时间的形状距离学习算法

形状距离学习是形状匹配框架中引入的后处理步骤, 能够有效改善逐对计算得到的形状间距离.利用期望首达时间分析形状间相似度可能导致距离更新不准确, 针对这一问题提出了一种基于广义期望首达时间 (Generalized mean first-passage time, GMFPT) 的形状距离学习方法.将形状样本集合视作状态空间, 广义期望首达时间表示质点由一个状态转移至指定状态集合所需的平均时间步长, 本文将其视作更新后的形状间距离.通过引入广义期望首达时间, 形状距离学习方法能够有效地分析上下文相关的形状相似度, 显式地挖掘样本空间流形中的最短路径, 并消除冗余上下文形状信息的影响.将所提出的方法应用到不同形状数据集中进行仿真实验, 本文方法比其他方法能够得到更准确的形状检索结果.     

A cross‐cultural study of the effect of a graph‐oriented computer‐assisted project‐based learning environment on middle school students' science knowledge and argumentation skills

The purpose of this mixed‐methods study was to explore how seventh graders in a suburban school in the United States and sixth graders in an urban school in Taiwan developed argumentation skills and science knowledge in a project‐based learning environment that incorporated a graph‐oriented, computer‐assisted application (GOCAA). A total of 42 students comprised the treatment condition and were engaged in a project‐based learning environment that incorporated a GOCAA. Of these 42 students, 21 were located in the United States and 21 were located in Taiwan. A total of 26 students comprised the control condition and were engaged in a project‐based learning environment without the GOCAA. Of these 26 students, 15 were in the United States and 11 were in Taiwan. In each country, verbal collaborative argumentation was recorded and the students' post‐essays were collected. A one‐way analysis of variance (ANOVA) was conducted for each measure of science knowledge about alternative energies. The results showed a significant treatment effect for the outcome of scientific explanation among U.S. students, while among Taiwanese students, a significant treatment effect on scientific facts was observed. A one‐way ANOVA was additionally conducted for each measure of argumentation skills and a significant treatment effect on counterarguments and rebuttals was observed among the U.S. students, while in Taiwan, a significant treatment effect on reasoning and rebuttals was observed. A qualitative analysis was conducted to examine how the GOCAA supported students' development of argumentation skills in different countries. This study found distinct argumentation patterns between the U.S. and Taiwanese intervention teams. Additionally, a distinct gender difference in the use of evidence and division of labour was noted when the Taiwanese teams were compared with the U.S. teams, which may be explained by cultural differences. This study concluded that, in both the United States and Taiwan, a project‐based learning environment incorporating a GOCAA was effective in improving students' science knowledge and developing their scientific argumentation skills.     

基于GPU的精确串匹配算法综述

精确 串匹配是计算机领域的一个经典问题。在大数据时代,海量的数据给串匹配问题带来巨大的挑战。当前,GPU的应用得到学术界和工业界的广泛关注。近年,基于GPU的串匹配算法研究已成为学术界的焦点。为展示近年的研究,本文综述了基于GPU的精确串匹配技术,针对不同的算法和GPU架构介绍精确串匹配技术在GPU上的改进：不同算法的改进具有差异性,研究时需扩展具体算法,并比较上述算法的优缺点。最后对评测指标进行介绍,展望其发展趋势。     

面向三维动画自动生成的路径规划

全过程计算机辅助自动生成动画技术由中科院陆汝钤院士于上世纪90年代提出,这一技术的目标是将适当的故事以受限自然语言的方式输入计算机,由开始一直到动画的生成,每一步都是在计算机的辅助下完成.在动画生成过程中,如何规划运动对象路径是影响动画效果的重要因素.为此,提出基于动画场景规划信息的路径规划方法,该方法通过预定义的路径定性规划语言PADL描述规划需求,然后通过利用扩展的A*算法规划路径并生成路径定量规划语言PCAL描述的路径,最后采用运动图方法合成路径运动动画.实验表明,该方法能有效地自动生成符合指定要求的路径动画.     

基于多特征融合和图匹配的维汉句子对齐

维吾尔语新闻网页与对应的中文翻译网页在内容上往往并非完全可比,主要表现为双语句子序列的错位甚至部分句子缺失,这给维汉句子对齐造成了困难。此外,作为新闻要素的人名地名很多是未登录词,这进一步增加了维汉句子对齐的难度。为了提高维汉词汇的匹配概率,作者自动提取中文人名、地名并翻译为维吾尔译名,构造双语名称映射表并加入维汉双语词典。然后用维文句中词典词对应的中文译词在中文句中进行串匹配,以避免中文分词错误,累计所有匹配词对得到双语句对的词汇互译率。最后融合数字、标点、长度特征计算双语句对的相似度。在所有双语句子相似度构成的矩阵上,使用图匹配算法寻找维汉平行句对,在900个句对上最高达到95.67%的维汉对齐准确率。     

基于案例推理的故障诊断算法

电力系统智能告警研究目前主要集中于规则推理领域,未能充分利用系统内历史故障的案例进行有效的分析与故障处理.本文初步研究了基于案例推理的故障诊断算法,在此基础上提出了一种基于向量计算的案例匹配算法.该算法通过对故障信息进行抽象,并与系统案例库中的案例进行相似度计算的方式,对故障信息与案例信息进行一一匹配.案例库的建立基于经过人工分析的各类历史故障信息.相似案例的故障性质、故障原因分析可对本次故障的分析、诊断和后续处理起到指导作用.     

基于联合特征匹配的多视角三维重建方法

本研究提出了一种基于角点与直线联合特征的三阶段匹配算法,依次经过相关匹配、松弛迭代匹配和最小平方中值法匹配三个步骤,并与此同时加入限定图像匹配区域、添加手工匹配点对以及局部直线匹配三个人机交互环节,将局部坐标系中得到的数据点进行三维数据融合到一个坐标系中,能较好解决视差不连续区域和遮挡区域的误匹配问题.实验证明该算法具有良好的运行效率和稳定性,能够对非平面物体进行精确三维重建并实现多视角显示.     

Imperfect premise matching controller design for T-S fuzzy systems under network environments

This paper focuses on an imperfect premise matching controller design for T-S fuzzy systems under network environments. Different with the traditional parallel distribution compensation (PDC) method, the same premises between the PDC controller and the T-S fuzzy systems are no longer needed again in the proposed method. Under consideration of the unmatched grades of membership in the networked T-S fuzzy systems, a unified T-S fuzzy model is firstly proposed, in which a networked state-feedback fuzzy controller with communication delays is used to reconstruct the system. Then, based on the constructed model and by use of the Wirtinger-based inequality technique to deal with the cross items, two less conservative stability and stabilization criteria are derived to enhance the design flexibility. Finally, two numerical examples are used to show the effectiveness of proposed method.     

基于Hu矩和纹理特征结合的人体异常行为识别

为了提高人体异常行为识别的准确率,采用了一种将多特征结合的异常行为识别算法,主要包括对步行、快跑、慢跑、拳击、双手挥舞、鼓掌六种异常行为进行识别。首先从视频流中提取出人体轮廓,然后从所得的轮廓中提取Hu矩特征与纹理特征。最后通过模板匹配的方法,采用马氏距离度量所需识别的当前行为特征向量与标准模板行为的特征向量之间的相似性,并通过设置相应的阈值判定该行为所属类别。实验证明,该方法比提取单一特征的方法识别率高,且具有一定的实用价值。