改进的稀疏子空间聚类算法

在现有的稀疏子空间聚类算法理论基础上提出一个改进的稀疏子空间聚类算法:迭代加权的稀疏子空间聚类。稀疏子空间聚类通过解决l1最小化算法并应用谱聚类把高维数据点聚类到不同的子空间,从而聚类数据。迭代加权的l1算法比传统的l1算法有更公平的惩罚值,平衡了数据数量级的影响。此算法应用到稀疏子空间聚类中,改进了传统稀疏子空间聚类对数据聚类的性能。仿真实验对Yale B人脸数据图像进行识别分类,得到了很好的聚类效果,证明了改进算法的优越性。     

自学习粒子群与梯度下降混杂的漏磁反演方法

利用自学习粒子群优化算法的全局寻优能力克服梯度下降法过分依赖初始解,易陷入局部极值的缺点,从而提高梯度下降法的优化性能。以径向基函数神经网络为前向模型,基于自学习粒子群与梯度下降混杂的反演方法用于漏磁缺陷轮廓重构中。实验结果表明,该反演方法重构的缺陷轮廓比较准确,且在漏磁信号存在噪声的情况下,重构结果到与实际轮廓相近,并具有一定的噪声鲁棒性。     

“What is dear to you?” Survey of beliefs regarding protection of critical infrastructure against terrorism

Up to the present, there is only very little research on how the population perceives terrorism and its threats, even though support from the population is crucial for effective counterterrorism. By eliciting beliefs and subjecting them to content analyses, six factors were found that determine the protection worthiness of a target in the people's view: the potential damage to “people,” “symbolism,” “economy,” “politics,” “nature,” and “image/publicity.” These empirically found factors are in line with factors specified by terrorist target selection models. They differ in the strength of their cognitive representation among participants and, thus, their subjective importance to the people. The first three factors are shared among all participants, whereas the latter ones could only be found in a part of the participant sample. People's judgments of the targets' protection worthiness differ substantially from their judgments of the targets' attractiveness to terrorists, even though the same factors seem to be involved. This study offers an insight into the people's mental model about protection worthiness of targets. Together with classical risk analysis and knowledge about terrorists, these results can form a basis for setting up a holistic scheme for critical infrastructure protection.     

用加权余量法求解弹性机构的稳态响应

在机构的运动弹性动力分析中,稳态响应的求解已有了多种方法。本文通过分析、比较,提出了用加权余量法求解弹性机构稳态响应的思想,并给出了两种基本算法。作为应用,本文还给出了两个算例,其中空间四杆RS'SR 机构为本文首次给出。     

交互式增广加权范数法及其应用

本文提出利用增广加权范数把多目标优化问题转化为标量优化问题,证明了转化后的问题与原问题完全等价,同时给出了目标函数值与权系数的关系,建立了以目标函数值作为交互媒介的多目标优化方法,最后给出了应用该方法的一个实例。     

基于小波的信噪分离方法

本文在 Ｄｏｎｏｈｏ 提出的从噪声中恢复信号的非线性小波方法的基本原理的基础上, 并针对非线性小波方法的核心问题——浮动阈值的设定提出了新的算法。通过算例证明, 所设计的浮动阈值在实际应用中效果较好, 特别在低信噪比时信号恢复效果明显优于现常用的浮动阈值设计方法。     

目标搜索的预测分析及搜索计划的优化逼近

给出了目标搜索的预测分析公式,提出了最优搜索计划的一种优化逼近方法及有关实例     

带交易费的市场模型和未定权益套期保值

给出了有交易费的且多种股票参与交易的市场模型,并且讨论在此市场模型下的未定权益套期保值问题,在这一模型中,假定仅债券与股票之间有交易,而不同股票之间不直接进行交易以及交易费率是常数。     

辅助鞅下投资消费资产模型的一些基本性质

针对所给出的有交易费的资产模型,引入了资产折算函数,并利用辅助鞅和凸函数对偶方法,讨论了该模型下折算资产优化的性质.