核最近邻凸包分类算法

为了增强最近邻凸包分类器的非线性分类能力,提出了基于核函数方法的最近邻凸包分类算法。该算法首先利用核函数方法将输入空间映射到高维特征空间,然后在高维特征空间采用最近邻凸包分类器对样本进行分类。最近邻凸包分类器是一类以测试点到各类别凸包的距离为相似性度量,并按最近邻原则归类的分类算法。人脸识别实验结果证实,这种核函数方法与最近邻凸包分类算法的融合是可行的和有效的。     

基于K最近邻算法的网络不良信息过滤系统研究

网络不良信息过滤的难点是如何对信息进行准确分类,而K最近邻分类器作为模式分类领域一种简单而有效的分类方法,在很多领域得到应用。提出了一种基于K最近邻算法的网络不良信息过滤方法,并对K最近邻算法计算量大的问题进行了一些改进。实验表明：采用K最近邻算法的不良信息过滤系统具有良好的应用效果。     

利用最近邻信息快速分类多标签数据

为克服ML-KNN在分类效率方面的局限性,提出了一种基于KNN的快速多标签数据分类算法FKMC,利用待分类实例的k个最近邻的局部信息进行排序分类。从已分类数据实例集中选择待分类数据实例的k个最近邻;根据每个最近邻拥有的标签数和每个标签归属的最近邻数对待分类实例进行排序分类。仿真结果表明,最近邻的选择方法对分类器性能有显著的影响;在分类效果上FKMC与ML-KNN相当,有时甚至优于后者;而在分类效率上FKMC则显著优于ML-KNN。     

基于遗传进化的最近邻聚类算法及其应用

提出了基于遗传进化的最近邻聚类算法，该算法结合了遗传算法（GA）与最近邻聚类算法（NN）。对要进行分类的样本和特征量进行优化选取，去除位于类交界的模糊样本，并对样本分类有效的特征量进行放大，对不利于样本分类的特征量进行抑制，从而提高了样本分类的精度，将该算法应用于抽水蓄能发电机组的工况分类，大大提高了机组工况的识别效果，验证了基于遗传算法的最近邻聚类算法的有效性。     

基于决策空间变换最近邻方法的Pareto支配性预测

为提高在决策空间运用最近邻方法预测多目标优化Pareto支配性的精度，提出一种基于决策空间变换的最近邻预测方法.在分析目标函数与决策分量相关性的基础上，提出属性变化趋势模型的构造方法，建立低计算成本的属性趋势代理模型.通过属性趋势模型引入决策空间到目标空间的映射知识，对多目标问题的决策空间进行变换，使决策空间的最近邻更有效反映目标空间的最近邻.选取具有不同相关系数特征的典型多目标优化问题，进行Pareto支配性预测的可对比实验，结果表明在新空间中运用最近邻方法可显著提高分类准确性.     

面向对象高分辨遥感影像分类研究

高空间分辨率遥感影像采用传统基于像元分类方法精度较低,本文通过分析高分辨遥感影像特征,采用面向对象的最近邻监督分类方法对QuickBird影像进行分类研究,首先对影像进行对象分割,然后将分割对象信息、形状特征与及上下文联系等特征构成特征空间进行最近邻监督分类,并与传统的基于像元最近邻分类方法分类进行比较分析,结果表明,本方法能够较好的识别高分辨率地物类型,总精度为92.19%,Kappa系数为0.8835,较好地改善分类效果,适合高分辨遥感影像分类。     

基于ICA和NFL与NN联合分类器的人脸识别

提出了一种基于最近邻特征线(NFL)与最近邻(NN)联合分类器进行人脸识别的方法。首先对人脸图像用主成分分析(PCA)降维,然后用快速独立变量分析(FastICA)提取独立基,分类时采用最近邻特征线和最近邻分类器的联合分类器进行分类。该方法综合了NFL和NN的优势,充分利用了同类之间相似,距离最短的性质。实验表明此方法提高了人脸识别率,是一种可行的人脸识别方法。     

一种基于预分类的高效最近邻分类器算法

本文的最近邻分类器算法是采用多分类器组合的方式对测试样本进行预分类，并根据预分类结果重新生成新的训练和测试样本集。对新的测试样本采用最近邻分类器进行分类识别，并将识别结果与预分类结果结合在一起进行正确率测试。在ORL人脸库上的实验结果说明，该算法对小样本数据的识别具有明显优势。     

SMwKnn:基于类别子空间距离加权的互k近邻算法

互k最近邻算法(mKnnc)是k最近邻分类算法(Knn)的一种改进算法,该算法用互k最近邻原则对训练样本以及k最近邻进行噪声消除,从而提高算法的分类效果。然而在利用互k最近邻原则进行噪声消除时,并没有将类别属性考虑进去,因此有可能把真实有效的数据当成噪声消除掉,从而影响分类效果。基于类别子空间距离加权的互k最近邻算法考虑到近邻的距离权重,既能消除冗余或无用属性对最近邻分类算法依赖的相似性度量的影响,又能较好地消除邻居中的噪声点。最后在UCI公共数据集上的实验结果验证了该算法的有效性。     

大数据下的快速kNN分类算法*

针对K最近邻算法测试复杂度至少为线性,导致其在大数据样本情况下的效率很低的问题,提出了一种应用于大数据下的快速KNN分类算法。该算法创新的在K最近邻算法中引入训练过程,即通过线性复杂度聚类方法对大数据样本进行分块,然后在测试过程中找出与待测样本距离最近的块,并将其作为新的训练样本进行K最近邻分类。这样的过程大幅度的减少了K最近邻算法的测试开销,使其能在大数据集中得以应用。实验表明,本文算法在与经典KNN分类准确率保持近似的情况下,分类的速度明显快于经典KNN算法。     

Operational estimation of primary production at large geographical scales

A protocol is developed for calculation of phytoplankton production from remotely-sensed data in the operational mode. The key element is an objective assignment, on a pixel-by-pixel basis, of the parameters required to implement a primary production model (parameters of the photosynthesis-response function and of the vertical distribution of pigment biomass). In a regional context, the assignment is made by searching the archived data on these parameters according to the (remotely-sensed) chlorophyll concentration and surface temperature. We refer to this approach as the Nearest-Neighbour Method. The procedure is justified on the basis of the known variation of bio-optical properties of phytoplankton with chlorophyll and temperature as well as through consideration of the seasonal variation of watercolumn stratification and its effect on the vertical pigment profile. We illustrate the method, and its justification, using data from the Northwest Atlantic Ocean. Using data from an oceanographic expedition not included in the archive, we find that the parameters estimated in this way are not significantly different from those obtained by direct measurement. We estimate the error associated with parameter assignment on the calculated phytoplankton production to be about 27%. Some potential limitations of the method are discussed.     

协作机器人的构型分析研究

针对协作机器人的构型分析研究,得出了偏置是影响机器人构型的一个重要因素。首先,对现有典型协作机器人的构型进行分析,通过构型间的内在联系与演化过程得到构型之间的差异。然后利用构型之间的差异引出了偏置的定义,并通过偏置对关节运动范围的影响将其分为S型偏置和Y型偏置。最后以全局性能指标、工作空间体积指标和灵活工作空间指标为性能评价标准,对iiwa、Sawyer、Yumi三种典型构型的性能进行了仿真。通过对比分析可知,偏置的存在使构型的全局性能和工作空间体积性能降低,但对灵活工作空间性能有提高作用。结果表明偏置是影响机器人性能的重要因素,该分析研究为协作机器人的构型设计提供了理论依据。     

群智能新研究:角色?匹配的狼群劳动分工

启发于生物狼群群体的劳动分工行为,本文提出一种新颖的角色?匹配狼群劳动分工方法。通过剖析自然狼群的生物学行为,概括提炼出狼群劳动分工行为的个体任务的专职化、个体角色可塑性和任务分配均衡性3个典型特征,并建立了生物狼群劳动分工行为与普适性任务分配问题之间的仿生映射关系;从狼群“个体?个体”+“个体?环境”的交互方式角度出发,分析了角色?匹配的狼群劳动分工的个体角色转换和任务调整机制,研究了狼群角色?匹配的柔性劳动分工机制,提出了一种新的群智能方法,即角色?匹配的狼群劳动分工方法;将狼群的角色?匹配劳动分工与蚁群的刺激?响应劳动分工和蜂群的激发?抑制劳动分工进行了比较分析,并展示出角色?匹配狼群劳动分工方法的应用前景。     

Expert system verification and validation: a survey and tutorial

Assuring the quality of an expert system is critical. A poor quality system may make costly errors resulting in considerable damage to the user or owner of the system, such as financial loss or human suffering. Hence verification and validation, methods and techniques aimed at ensuring quality, are fundamentally important. This paper surveys the issues, methods and techniques for verifying and validating expert systems. Approaches to defining the quality of a system are discussed, drawing upon work in both computing and the model building disciplines, which leads to definitions of verification and validation and the associated concepts of credibility, assessment and evaluation. An approach to verification based upon the detection of anomalies is presented, and related to the concepts of consistency, completeness, correctness and redundancy. Automated tools for expert system verification are reviewed. Considerable attention is then given to the issues in structuring the validation process, particularly the establishment of the criteria by which the system is judged, the need to maintain objectivity, and the concept of reliability. This is followed by a review of validation methods for validating both the components of a system and the system as a whole, and includes examples of some useful statistical methods. Management of the verification and validation process is then considered, and it is seen that the location of methods for verification and validation in the development life-cycle is of prime importance.     

A method of defuzzification based on the approach of areas' ratio

The new method of defuzzification of output parameters from the base of fuzzy rules for a Mamdani fuzzy controller is given in the paper. The peculiarity of the method is the usage of the universal equation for the area computation of the geometric shapes. During the realization of fuzzy inference linguistic terms, the structure changes from the triangular into a trapezoidal shape. That is why the universal equation is used. The method is limited and can be used only for the triangular and trapezoidal membership functions. Gaussian functions can also be used while modifying the proposed method. Traditional defuzzification models such as Middle of Maxima − MoM, First of Maxima − FoM, Last of Maxima − LoM, First of Suppport − FoS, Last of Support − LoS, Middle of Support − MoS, Center of Sums − CoS, Model of Height − MoH have a number of systematic errors: curse of dimensionality, partition of unity condition and absence of additivity. The above-mentioned methods can be seen as Center of Gravity − CoG, which has the same errors. These errors lead to the fact that accuracy of fuzzy systems decreases, because during the training root mean square error increases. One of the reasons that provokes the errors is that some of the activated fuzzy rules are excluded from the fuzzy inference. It is also possible to increase the accuracy of the fuzzy system through properties of continuity. The proposed method guarantees fulfilling of the property of continuity, as the intersection point of the adjustment linguistic terms equals 0.5 when a parametrized membership function is used. The causes of errors and a way to delete them are reviewed in the paper. The proposed method excludes errors which are inherent to the traditional and non- traditional models of defuzzification. Comparative analysis of the proposed method of defuzzification with traditional and non-traditional models shows its effectiveness.     

并行模糊测试综述

软件脆弱性已成为互联网安全的主要威胁来源,软件脆弱性分析技术的重要性日益突出。模糊测试是脆弱性分析的热点技术之一,通过持续生成测试用例、动态监控目标代码执行和反馈调节变异策略的方法尝试触发程序异常,具有部署便捷、适用性广和效果直观的优点。随着测试目标的复杂性增加,从业人员对模糊测试的效率提出了更高的要求。并行模糊测试通过并行执行、任务分解和共享信息等方法提高脆弱性分析的效率。首先,分析了基于覆盖反馈的模糊测试面临的主要挑战;之后,探讨了并行模糊测试的解决思路和方案,从系统结构、任务划分、语料库共享和崩溃去重等方面对并行模糊测试进行了综述;最后,总结了现有并行模糊测试的优缺点,并对未来发展方向进行了展望。     

构建安全的局域网策略

网络安全是一门涉及计算机科学、网络技术、密码技术、信息安全技术、应用数学、教论、信论等多种学科的综合性学科,它涉及的因素主要包括物理安全、系统安全、信息安全和文化安全等范围。局域网络中人为的、非人为的,有意的、无意的;非法用户入侵、计算机病毒、黑客的不断侵袭等随时威胁局域网的安全。随着计算机网络技术迅速发展,网络安全问题也日渐突出。该文分析了威胁局域网安全的主要因素,探讨了从物理环境、网络层、应用层以及管理制度方面,运用VPN、NAT、VLAN、IDS、身份验证等技术构建安全局域网的策略。     

Remote sensing of spring phenology in boreal regions: A free of snow-effect method using NOAA-AVHRR and SPOT-VGT data (1982-2004)

Measurements of spring phenological dates in boreal regions using NDVI can be affected by snowmelt. This impacts the analysis of interannual variations in phenology and the estimates of annual carbon fluxes. For these two objectives, snowmelt effect must be removed from the phenological detection. We propose a methodology for determining the date of onset of greening in the 1982-2004 period using SPOT-VEGETATION (VGT) and NOAA Advanced Very High Resolution Radiometer (AVHRR) data. From 1998 onwards, the date of onset of greening is taken as the date at which the Normalized Difference Water Index (NDWI), calculated from SPOT-VGT near and short-wave infrared bands, starts increasing. This index decreases with snowmelt but increases with vegetation greening. For the 1982-2001 period, the date of onset of greening is the date at which AVHRR-NDVI equals a pixel specific threshold (PST), determined using the results of the NDWI method in the years common to the two datasets. The methods are validated using in situ measurements of the dates of leaf appearance. RMSE of 6.7 and 7.8 days, respectively, is found using NDWI-VGT and PST-NOAA methodologies, and the difference between the two methodologies in the common years is small. Very importantly, the dates are not biased. The interannual variations of the 23-year spring phenology dataset on the study area in northern Eurasia are analysed. In average over the study area, an advance of 8 days and a delay of 3.6 days are, respectively, found over the periods 1982-1991 and 1993-2004. These results confirm and complete previous studies about the greening trend, remove the uncertainty due to snow, and may improve carbon budget calculations.     

上下文感知旅游推荐系统研究综述

随着人们生活水平的提高，旅游已成为一项普遍的休闲活动，进而推动了旅游推荐方面技术的研究。与传统推荐系统相比，除了考虑游客和旅游产品的相关特征之外，旅游推荐系统的推荐质量在很大程度上受到位置、时间、天气、游客社交群体等上下文信息的影响。本文首先给出上下文感知旅游推荐系统的总体框架；然后对位置、时间、游客社会化网络和多维上下文等4类典型的上下文信息在旅游推荐系统中的应用进行了详细考察，并对综合应用各种上下文信息的旅游推荐系统进行了分析；从旅游推荐产品的角度对推荐系统进行分类考察；最后讨论了上下文感知旅游推荐系统目前面临的重点和难点问题，指出下一步的研究方向。     

Overview of ATM networks: functions and procedures

The purpose of this paper is to present an overview of the ATM technique. The basic principles of the ATM transfer technique, the B-ISDN layered protocol reference model, as well as the architecture and the main functions of the ATM network, are described in this tutorial. In particular, we present the main characteristics of the ATM technique, the structure of the ATM cell, the functions which deal with the protection from errored cells (HEC procedure) and the identification of the boundaries of the ATM cells (cell delineation and scrambling). The VP/VC concept is explained in detail, and the main principles and functions of the three lowest layers (Physical, ATM and Adaptation) of the B-ISDN layered protocol reference model are surveyed. The issue of Quality of Service parameters in ATM networks is reviewed, and the problem of the allocation of the network resources in response to user demands, as well as the problems of congestion control, Usage Parameter Control (UPC) and traffic modelling in ATM networks, are assessed. Finally, we present ‘possible’ sets of traffic characterization parameters, and show how possible B-ISDN sources can be modelled according to the basic on-off ATM traffic source model.