项目名称: 协方差阵的推断及在方向数据分析中的应用
项目编号: No.11471264
项目类型: 面上项目
立项/批准年度: 2015
项目学科: 数理科学和化学
项目作者: 马铁丰
作者单位: 西南财经大学
项目金额: 50万元
中文摘要: 本项目着力解决协方差阵的压缩估计以及方向数据的相关分析和多元统计建模等问题。在协方差阵的估计方面,采用矩阵分块技术和分解技术相结合的处理手段,建立新的压缩估计方法,并给出一些优良性的评价结果;在方向数据的统计建模方面,针对方向数据的结构特点,在本项目前半部分关于协方差阵压缩估计研究成果的基础上提出一些方向数据协方差阵的定义方法,并在其基础上建立更合理、更合适的回归模型,以解决目前方向数据回归模型与相关度量间相脱离的问题。为避开贝赛尔函数给似然推断带来的迭代算法耗时长和不稳健等困难,基于前半部分得到的协方差阵估计方面的研究成果,本项目为方向数据多元模型的参数推断提供更加快速和稳健的新方法。
中文关键词: 多元统计分析;协方差阵;高维数据分析;方向数据;统计推断
英文摘要: The objective of this project is to solve some problems on shrinkage estimation of the covariance matrices and to discuss correlation analysis and multivariate modeling for directional data. In terms of the estimation of covariance matrices, the combination of block and decomposition is used to create some new improved estimators and study their statistical properties. According to the structural characteristics of the directional data, some new definitions are proposed to product the covariance matrix based on the front research on shrinkage estimation of covariance matrix, and more reasonable and appropriate regression models follows these definitions to deal with the disconnect between correlation and regression at present. To avoid the time-consuming and unrobustness of iterative algorithm caused by Bessel function when the likelihood inference is considered, based on the above research results for covariance matrix, this object propose some new fast and robust methods for modeling multivariate directional data.
英文关键词: multivariate statistical analysis;covariance matrix;High-dimensional data analysis;directional data;statistical inference