基于特征子空间邻域的局部保持流形学习算法*

局部保持流形学习算法通过保持局部邻域特性来挖掘隐藏在高维数据中的内在流形结构。然而,对于缺乏足够训练样本的高维数据集,或者高维数据集存在非线性结构和高维数据特征中存在冗余、干扰特征,使得在原特征空间中利用欧式距离定义的邻域关系并不能真实反映数据的内在流形结构,从而影响算法的性能。提出利用正约束寻找特征子空间的方法,使得在此子空间中更多的同类样本紧聚,并进一步在该子空间中构建邻域关系来挖掘高维数据的内在流形,形成基于特征子空间邻域特性的局部保持流形学习算法(NFS-LPP和NFS-NPE)。它们在一定程度上克服了高维小样本数据集难以正确挖掘内在流形结构的问题,在Yale和ORL人脸库上的分类和聚类实验验证了其有效性。     

基于位置敏感哈希的网络视频重复检测

针对当前网络上存在着大量的重复或近似重复的视频问题,提出了一种基于镜头层比较和位置敏感哈希的快速准确的网络视频重复检测方法。通过视频间匹配的镜头数占查询视频总镜头数的比例来判断视频的相似性。除此之外,还利用著名的近似最近邻查找技术——LSH在镜头层来快速查找相似镜头,从而提高检测速度。通过将镜头作为检索单元,把数据库中所有视频的镜头放到一起构建一个新的数据集,将种子(查询)视频的每一个镜头作为一个查询请求,应用基于LSH的近似近邻检索方法,检索出与查询镜头相匹配的所有镜头,最后融合这些返回的结果,得到查询视频的重复或者近似重复的视频集。通过在包含12 790个视频的CC_WEB_VIDEO数据集上的实验结果表明,该方法取得了相比已有方法更好的检测性能。     

面向情感分析的短文本意义串发现及分析算法

针对网络环境中大量短文本信息的鲜明特点,突破传统基于词标注的分类方法,提出了一种短文本频繁模式发现及其有效意义串分析算法。通过改进的FP-树算法挖掘最大频繁模式得到关键词库之后,结合中文词语局部性原理对关键词按照位置点聚类,进行意义串二次挖掘,最后进行文本情感分析。仿真结果表明,对于意义串词组挖掘,该算法具有较高的准确率,并有助于及时了解网络群体的情感方向。     

提高粗酚产率的措施

针对焦油加工系统粗酚产率低的问题,分析了其原因,探讨了解决的措施,通过稳定馏分塔的塔顶温度,将酚、萘、洗三混馏分的初馏点控制在190～200℃,大幅度提高和稳定了酚的集中度,并对馏分塔回流、洗涤脱酚、分解操作进行优化和改进,减少了分解过程中酚的流失,提高了酚的脱除率,粗酚产率大幅度提高。     

500kV悬挂式管型母线施工方法改进

本文对改良焊接管母导轨、改进管母的吊装方法采用同步到位，降低劳动强度，提高劳动效率进行了讨论，可供遇有相同或类似的施工进行参考。     

一种利用数据融合来提高局部性和减少伪共享的方法

某些应用程序不能通过数组内元素的重排优化获得性能提高 .针对这一问题 ,该文扩展了数组之间数据重组优化方法 ,着重分析了将多个数组的数据按一定方式进行融合来提高局部性和减少伪共享优化方法的特性 .文章针对几种典型的数组关联模式 ,提出了相应的数据融合方法 ,并建立了一组粗略的性能代价判别规则 ,以指导编译器有选择地融合数组以提高程序的全局优化效果 .根据在多个平台上的测试结果 ,该文还分析了数据融合优化方法在不同体系结构上的性能可移植性 ,并将体系结构特征加入到性能代价判别规则中 ,使得此优化方法能适用于不同的体系结构 .测试结果表明 ,数据融合优化方法对提高某些应用程序的性能 ,尤其是其在软件DSM体系结构上的性能 ,是非常有效的     

Personalising learning through the use of technology

In March 2005, the Department for Education and Skills published its e-strategy, Harnessing Technology (DfES, 2005, Harnessing technology: Transforming Learning and Children’s Services). Within this, two of its key objectives were: firstly to transform teaching and learning, and help to improve outcomes for children and young people, through shared ideas, more exciting lessons, and online help for professionals. Secondly, to engage ‘hard to reach’ learners, with special needs support, in more motivating ways of learning, and give them more choice about how and when they learn.     

The impact of online video lecture recordings and automated feedback on student performance

To what extent a blended learning configuration of face-to-face lectures, online on-demand video recordings of the face-to-face lectures and the offering of online quizzes with appropriate feedback has an additional positive impact on the performance of these students compared to the traditional face-to-face course approach? In a between-subjects design in which students were randomly assigned to a group having access to the online lectures including multiple choice quizzes and appropriate feedback or to a group having access to the online lectures only, 474 students (161 men and 313 women) of a course on European Law agreed to participate in the experiment. By using regression analysis we found that the course grade of the students was predicted by their grade point average, their study discipline, their grade goal for the course, the expected difficulty-level of the course, the number of online lectures they viewed, the number of lectures the students attended in person and the interaction between the lectures they viewed online and attended in person. Students who attended few lectures had more benefit from viewing online lectures than students who attended many lectures. In contrast to our expectations, the regression analysis did not show a significant effect of automated feedback on student performance. Offering recordings of face-to-face lectures is an easy extension of a traditional course and is of practical importance, because it enables students who are often absent from the regular face-to-face lectures to be able to improve their course grade by viewing the lectures online.     

Smart utilization of tertiary instructional modes

This empirical research surveys first year tertiary business students across different campuses regarding their perceived views concerning traditional, blended and flexible instructional approaches. A structural equation modeling approach shows traditional instructional modes deliver lower levels of student-perceived learning quality, learning experience and learning skills. A combination of on-line and face-to-face learning approaches, embedded across each course, yields far higher levels of total learning effects, and to explain differences in instructional approaches, a ‘Cone of Learning’ continuum is presented and discussed. Theoretical and practical research implications, and the measurement, theoretical and management aspects of future research options are presented. Tertiary institutions can adopt the approaches herein to assist in the development and build of smart targeted learning solutions – ones more in-line with the perceived needs of their respective student year levels and groups.     

Web访问特征模型建模

Web访问特征模型建模是进行有效Web缓存管理的基础。该文根据Web访问的四个典型特征建立综合的数学模型,实现了Web访问特征建模生成器(WebGenM),实验表明模拟器能较好地模拟网络访问流的特征,而且易于使用,具有较大的灵活性,其为进一步的Web缓存和预取技术的研究提供了重要依据。