基于线性规划支持向量回归的混沌系统预测

支持向量机是一种基于统计学习理论的新颖的机器学习方法,该方法已经广泛用于解决分类与回归问题。标准的支持向量机算法需要解一个二次规划问题,当训练样本较多时,其运算速度一般很慢。为了提高运算速度,介绍了一种基于线性规划的支持向量回归算法,并由此提出几种新的回归模型,同时将它们应用到混沌时间序列预测中,并比较了它们的预测性能。在实际应用中,可以根据具体情况灵活地选择所需模型。     

基于SVM的大样本数据回归预测改进算法

针对支持向量机回归预测精度与训练样本尺寸不成正比的问题,结合支持向量机分类与回归算法,提出一种大样本数据分类回归预测改进算法。设计训练样本尺寸寻优算法,根据先验知识对样本数据进行人为分类,训练分类模型,基于支持向量机得到各类别样本的回归预测模型,并对数据进行预测。使用上证指数的数据进行实验,结果表明,支持向量机先分类再回归算法预测得到的均方误差达到12.4,低于人工神经网络预测得到的47.8,更远低于支持向量机直接回归预测得到的436.9,验证了该方法的有效性和可行性。     

基于AdaBoost算法的加权二乘向量回归机

针对二乘向量机(LS-SVM)对所有样本误差惩罚相同、预测精度不高的问题,提出了一种基于AdaBoost模型的二乘向量回归机。该算法使用多个二乘向量机按照某种学习规则协调各二乘向量机的输出,同时根据回归精度,建立各二乘向量机中每一个样本的误差惩罚权重,以突出样本的惩罚差异性,提高算法的泛化性能。实验结果表明,提出的算法提高了二乘向量回归机的预测精度,优化了学习机的性能。     

基于迁移子空间学习的偏最小二乘回归软测量方法

针对流程工业中工况改变易导致当前样本与历史样本分布失配,传统软测量模型失准的问题,考虑工业数据时序性、动态性以及存在过程漂移等特性对建模的影响,提出一种基于迁移子空间学习的偏最小二乘回归软测量方法.首先,回归框架采用非线性迭代偏最小二乘方法,对其求解映射向量的目标函数施加基于子空间重构的域适应正则项,映射过程中保证当前工况中每个样本能够被历史工况样本线性重构.在此基础上对重构矩阵施加低秩稀疏约束,保持数据结构的同时使重构矩阵具备块状结构以应对过程漂移特性.将所提出方法在1个数值案例和3个不同的多工况数据集中进行实验,并与现有域适应回归方法进行对比分析.实验表明,所提出方法能够有效提高模型在跨工况条件下的预测精度,减少工况间数据分布差异对模型性能的影响.     

融合概率分布和单调性的支持向量回归算法

传统支持向量回归是单纯基于样本数据的输入输出值建模,仅使用样本数据信息,未充分利用其他已知信息,模型泛化能力不强.为了进一步提高其性能,提出一种融合概率分布和单调性先验知识的支持向量回归算法.首先将对偶二次规划问题简化为线性规划问题,在求解时,加入与拉格朗日乘子相关的单调性约束条件;通过粒子群算法优化惩罚参数和核参数,优化目标包括四阶矩估计表示的输出样本概率分布特性.实验结果表明,融合这两部分信息的模型,能使预测值较好地满足训练样本隐含的概率分布特性及已知的单调性,既提高了预测精度,又增加了模型的可解释性.     

基于Help-Training 的半监督支持向量回归

提出一种基于Help-Training的半监督支持向量回归算法,包含最小二乘支持向量回归（LS-SVR）和近邻（NN）两种类型学习器.主学习器LS-SVR通过选择高置信度的未标记样本加以标记,并将其添加到已标记样本集,使训练样本的规模不断扩大,以提高LS-SVR的函数逼近性能.辅学习器NN用以协助LS-SVR从训练样本比较密集的区域选取未标记样本加以置信度评估,可以减弱噪声对学习效果的负面影响.实验结果表明所提算法具有良好的回归估计性能,学习精度较高.     

基于最小二乘支持向量回归的水质预测

水质系统是一个开放的、复杂的、非线性动力学系统,具有时变复杂性,针对水质预测方法的研究虽然已经取得了一些成果,但也存在预测精度与计算复杂度等难题。为此,本文提出一种基于最小二乘支持向量回归的水质预测算法。支持向量机是机器学习中一种常用的分类模型,通过核函数将非线性数据从低维映射到高维空间,在高维空间实现线性分类和回归,最小二乘支持向量回归（LS-SVR）利用所有的样本参与回归拟合,使得回归的损失函数不再只与小部分支持向量样本有关,而是由所有样本参与学习修正误差,提高预测精度;同时该算法将标准SVR求解问题由不等式的约束条件及凸二次规划问题转化成线性方程组来求解,提高了运算速度,解决了非线性复杂特性的水质预测问题。     

基于二次学习的半监督字典学习软件缺陷预测*

当软件历史仓库中有标记训练样本较少时,有效的预测模型难以构建.针对此问题,文中提出基于二次学习的半监督字典学习软件缺陷预测方法.在第一阶段的学习中,利用稀疏表示分类器将大量无标记样本通过概率软标记标注扩充至有标记训练样本集中.再在扩充后的训练集上进行第二阶段的鉴别字典学习,最后在学得的字典上预测缺陷倾向性.在NASA MDP和PROMISE AR数据集上的实验验证文中方法的优越性.     

基于多样性加权相似度的集成局部加权偏最小二乘软测量建模

针对传统即时学习软测量方法仅考虑单一的相似度函数,难以有效处理复杂工业过程中的非线性特性,从而导致模型预测性能受限的问题,提出了一种基于多样性加权相似度（DWS）的集成局部加权偏最小二乘（LWPLS）软测量建模方法.首先采用随机子空间法和高斯混合聚类,构建一组多样性的训练样本子集;然后通过偏最小二乘回归分析确定输入特征权值,从而定义一组多样性加权相似度函数.在线实施阶段,对于任意的查询样本,基于多样性的相似度指标,可建立一组多样性的LWPLS软测量模型,随后引入集成学习策略实现难测变量的融合预测.在数值例子和脱丁烷塔过程中的应用结果表明了该方法的有效性.     

基于灰色-RBF神经网络的传播损耗模型训练

室内信号传播损耗模型是基于信号强度测距法的射频识别定位技术的关键。但因室内环境较为复杂且受到多径效应等因素影响,传统的基于经验的信号传播损耗模型环境适应性差,导致测距定位误差较大;而利用传统的神经网络进行传播损耗模型训练则存在所需训练样本过多、硬件采集工作量大等缺点。针对以上问题,提出在变密度采样模式下的基于灰色理论与RBF神经网络相结合的传播损耗模型训练方法。基于灰色理论,利用少量样本预测得到更多样本,并与部分原始样本共同重组样本数据进行RBF网络的训练,以构建传播损耗模型。实验结果表明,该方法可以利用较少的训练样本准确地建立室内信号传播损耗模型,可以很好地满足室内测距定位的精度要求,并可大大减少样本采集工作量。     

自动化回归测试的技术和实现

提出了两种提高回归测试自动化程度的技术。其中一种技术采用数据驱动的方式,使得测试脚本成为可以驱动所有类似测试用例组的通用脚本,同时,实现了测试执行和测试逻辑的分离,使得测试用例的修改和维护更加容易。介绍的另一种技术使用附加的动态链接库来恢复被测软件的图形界面状态,使得软件图形界面的自动测试不易受到被测软件状态改变的影响,提高了整个自动测试系统的健壮性。     

Continuous Software Process Improvement Requires Organisational Learning: An Australian Case Study

The study reported in this paper suggests that in order to achieve optimal benefits from implementing process improvement programs, organisations must move towards becoming what is termed a learning organisation. Software process assessment leads to the identification and selection of key activities for improvement and the continuous application of improvements to match business needs (ISO/IEC 1996). Continuous improvement requires a commitment to learning on the part of the organisation (Garvin 1993). A model to help identify evidence of learning (the Organisational Learning Evaluation Cycle [OLEC] has been developed and empirically tested in the study. We have found evidence to suggest that the case study organisation had not moved through all three of Garvin's (1993) overlapping phases of organisational learning and as a result the firm's improvement program did not achieve optimal benefits for the organisation. The paper concludes by discussing why significant improvement in performance was not achieved.     

基于标记分布学习的异态集成学习算法

为了提高预测的准确性,文中结合机器学习中堆积(Stacking)集成框架,组合多个分类器对标记分布进行学习,提出基于标记分布学习的异态集成学习算法(HELA-LDL).算法构造两层模型框架,通过第一层结构将样本数据采用组合方式进行异态集成学习,融合各分类器的学习结果,将融合结果输入到第二层分类器,预测结果是带有置信度的标记分布.在专用数据集上的对比实验表明,HELA-LDL可以发挥各种算法在不同场景下的性能较优,稳定性分析进一步说明算法的有效性.     

基于半监督集成学习的软件缺陷预测*

在软件缺陷预测中,标记样本不足与类不平衡问题会影响预测结果.为了解决这些问题,文中提出基于半监督集成学习的软件缺陷预测方法.该方法利用大量存在的未标记样本进行学习,得到较好的分类器,同时能集成一系列弱分类器,减少多数类数据对预测产生的偏倚.考虑到预测风险成本问题,文中还采用训练样本集权重向量更新策略,降低有缺陷模块预测为无缺陷模块的风险.在NASA MDP数据集上的对比实验表明,文中方法具有较好的预测效果.     

参数依赖型软件参数更动测试策略和方法

参数依赖型软件是指初始化时读取并解析配置参数,并据此进行任务处理的软件,航天测控软件是典型的参数依赖型软件。航天测控软件具有明显的领域软件特征,多采用领域工程分析技术,实现业务处理逻辑和具体任务参数的分离,达到仅通过修改任务配置参数而适应高强度型号任务的目的。通过对参数依赖型软件架构、应用模式的分析,提出一种对参数依赖特性进行验收测试、参数更动测试的流程、策略和方法。并基于该方法,对远程数据交互软件进行了参数依赖特性测试,测试结果表明,该方法具有测试覆盖性强、测试重点突出、测试效率高的特点。     

Visual comparison of software cost estimation models by regression error characteristic analysis

The well-balanced management of a software project is a critical task accomplished at the early stages of the development process. Due to this requirement, a wide variety of prediction methods has been introduced in order to identify the best strategy for software cost estimation. The selection of the best technique is usually based on measures of error whereas in more recent studies researchers use formal statistical procedures. The former approach can lead to unstable and erroneous results due to the existence of outlying points whereas the latter cannot be easily presented to non-experts and has to be carried out by an expert with statistical background. In this paper, we introduce the regression error characteristic (REC) analysis, a powerful visualization tool with interesting geometrical properties, in order to validate and compare different prediction models easily, by a simple inspection of a graph. Moreover, we propose a formal framework covering different aspects of the estimation process such as the calibration of the prediction methodology, the identification of factors that affect the error, the investigation of errors on certain ranges of the actual cost and the examination of the distribution of the cost for certain errors. Application of REC analysis to the ISBSG10 dataset for comparing estimation by analogy and linear regression illustrates the benefits and the significant information obtained.     

Are learning styles useful indicators to discover how students use Scrum for the first time?

Teaching agile practices is in the cutting-edge of Software Engineering education since agile methodologies are widely used in the industry. An effective strategy to teach agile practices is the use of a capstone project, in which students develop requirements following an agile methodology. To improve students’ learning experience, professors have to keep track and analyze the information generated by the students during the capstone project development. The problem here arises from the large amount of information generated in the learning process, which hinders professors to meet each student’s learning profile. Particularly, to know the students skills and preferences are key aspects on a learner-centered approach of education in order to personalize the teaching. In this work, we aim to discover the relationships between students’ performance along a Scrum-based capstone project and their learning style according to the Felder–Silverman model, towards a first step to build the profiles. To address this issue, we mined association rules from the interaction of 33 Software Engineering students with Virtual Scrum, a tool that supports the development of the capstone project in the course. In the present work we describe promising results in experiments with a case-study.     

基于连续型深度置信神经网络的软件可靠性预测

为了提高软件可靠性智能预测的精度,采用连续型深度置信神经网络算法用于软件可靠性预测。首先提取影响软件可靠性的核心要素样本,并获取样本要素的关键特征;然后建立连续型深度置信神经网络(Deep Belief Network,DBN)的软件可靠性预测模型,输入待预测样本,通过多个受限波尔兹曼机(Restricted Boltzmann Machine,RBM)层的预处理训练,以及多次反向微调迭代获取DBN权重等参数,直到达到最大RBM层数和最大反向微调迭代次数;最后获得稳定的软件可靠性预测模型。实验结果证明,通过合理设置DBN隐藏层节点数和学习速率,可以获得良好的软件可靠性预测准确率和标准差。与常用的软件可靠性预测算法相比,所提算法的预测准确度高且标准差小,在软件可靠性预测方面的适用度较高。     

Learning performance and computer software: an exploration of knowledge transfer

Computer studies educators have a challenging task in keeping pace with the rapidly changing content of computer software. One way to meet this challenge is to examine the nature of knowledge transfer. Instead of focusing on unique software packages, teachers could concentrate on knowledge that is likely to transfer from one software application to another. The purpose of the current study was to describe what kind of knowledge is used in learning new software, assess the relative effectiveness of this knowledge in aiding the learning process, and examine how the results could advance educational learning theory and practice. Thirty-six adults (18 male, 18 female), representing three computer ability levels (beginner, intermediate, and advanced), volunteered to think out loud while they learned the rudimentary steps (moving the cursor, using a menu, entering data) required to use a spreadsheet software package (Lotus 1-2-3). Previous understanding of terminology, software concepts and actions, and other software packages had the largest impact, both positive and negative, on learning. A basic understanding of the keyboard and common movement keys was also important, although higher level knowledge (e.g., terms, concepts, actions) is probably necessary for significant gains in learning performance. Computer ability had little impact on the type of transfer knowledge used, except with respect to the use of software concepts and, to a lesser extent, terminology. The interaction between problem type and effectiveness of a specific transfer area suggests that identifying specific common tasks among software packages is important in detecting useful transfer knowledge. It is equally important that computer users understand labeling idiosyncrasies of these common tasks.