光流算法研究

光流的测量是视频序列处理和视频挖掘中的基本问题。首先简述了光流法的基本原理,即物体亮度不变特性;然后阐述了计算光流的4类算法:微分方法、区域匹配方法、基于能量的方法和基于相位的方法,并对当前主流的能量算法进行了详细探讨;最后对各种光流算法的评测方法与测试集进行了说明。     

THE HARD CASES

Many countries received Soviet-origin highly enriched uranium (HEU) for civilian nuclear research purposes. Because of inadequate nuclear security at a number of the research sites, U.S. policy has sought to remove or otherwise safely dispose of their HEU stocks as quickly as possible. Although the pace of HEU disposition has accelerated significantly in recent years, several sites have posed formidable technical, economic, and political challenges. This article identifies the major obstacles to HEU removal at two key installations—Kharkiv in Ukraine, and Sosny in Belarus—and recommends a strategy for overcoming these impediments. Key components for a successful disposition strategy include: treating these cases with the urgency they deserve, expanding potential compensation packages, explicitly addressing the institutional and political issues involved, engaging high-level political leaders, working with third parties, and promoting these efforts as part of a nondiscriminatory initiative to phase out HEU in the civilian nuclear sector globally.     

Confronting the “Perpetual Menace to Human Security”

Nuclear weapon states historically have attached great secrecy to their nuclear weapon and fissile material production programs and stockpiles, despite warnings that this would fuel fears, handicap informed debate and decision making, and drive arms races. As evidenced by the “Action Plan on Nuclear Disarmament” agreed upon at the 2010 Treaty on the Non-Proliferation of Nuclear Weapons (NPT) Review Conference, however, the international community now sees greater transparency about nuclear weapon and fissile material stocks as necessary for enabling and monitoring progress toward nuclear disarmament. To support this effort, the International Panel on Fissile Materials has proposed a step-by-step program for weapon states to declare their inventories, production histories, and disposition of nuclear warheads and fissile materials, and to set up joint projects to develop methods for verifying these declarations. This openness initiative is described here, and could be adopted at the 2015 NPT Review Conference, laying a basis for negotiating verifiable deep reductions in nuclear arsenals and their eventual elimination.     

基于数据挖掘的网络化教学研究

基于传统网络化教学系统存在的问题,应用数据挖掘的相关技术,对以往在网络化教学系统中不能对学生练习结果进行科学、客观分析的实际问题给出了初步解决方案。基于数据挖掘技术的网络化教学系统,其目的就是在一定程度上提高教师的教学效率。．     

基于FCM与粗糙集的连续数据知识挖掘方法

对连续数据知识挖掘可采用模糊C均值聚类将其离散化,再应用粗糙集进行约简.文中针对该过程中样本数据离散化后出现多个聚类中心隶属度值较接近的情况,提出了隶属度重叠度的方法来扩展对应离散类别的选取.离散类别确定后再应用粗糙集理论对其进行知识挖掘,可得到连续数据的本质特性.此方法应用于汽轮机轴系振动数据的知识挖掘.与采用最大隶属度确定对应类别的方法相比,该方法不影响对连续数据的知识挖掘,并能真实反映连续数据的特性.     

REFASHIONING AUSTRALIA'S NUCLEAR BARGAIN?

This article explores the challenges that Australia faces in reconciling its commitments to nonproliferation and uranium exports during a time when the international nuclear nonproliferation regime is under major stress and the world uranium market is bullish. The “grand bargain” that has framed Australian participation in the nonproliferation regime and the nuclear fuel market since the 1970s was only tenable in an era of stagnant uranium demand and a stable nuclear balance. However, contemporary nuclear proliferation dynamics and the revival of interest in nuclear energy have accentuated the incompatibility between Australia's commitment to nonproliferation and the desire to profit from uranium exports. The contemporary international strategic environment, international nonproliferation regime, and nuclear energy market are characterized by developments that not only undermine the basis of Australia's grand bargain, but also present challenges and opportunities for the refashioning of Australian policy.     

数据挖掘在图像信息隐写分析中的应用

提出数据挖掘在图像信息隐写分析中的应用。首先计算原始图像和隐写图像在通用图像质量指标下的失真测度,然后利用数据挖掘技术建立一个预测模型,完成失真测度到已知隐写概率的映射,模型的输出即为预测隐写概率。实验结果表明,使用该方法得到的预测隐写概率和已知隐写概率的相关度,当使用训练样本时达到0.951以上,使用测试样本时达到0.943以上,达到了很好的预测效果。     

多媒体数据挖掘中数据间的相似性度量研究

聚类是多媒体数据挖掘的重要任务之一,数据之间的相似性度量是聚类的基础和前提。多媒体数据的特征矢量通常都是数十维甚至数百维的,但传统的相似度量方式一般只适用于低维数据。在分析高维数据特性的基础上,提出了一个新的度量方式。通过使用一个特定的策略对原始数据空间进行网格划分,该方法较好地避免了噪声数据对高维数据相似性度量的影响。实验证明此方法是有效的。