项目名称: 高维数据的非参数经验贝叶斯方法
项目编号: No.11201327
项目类型: 青年科学基金项目
立项/批准年度: 2013
项目学科: 数理科学和化学
项目作者: 姜文华
作者单位: 苏州大学
项目金额: 22万元
中文摘要: 本项目旨在发展高维数据中的非参数经验贝叶斯方法。高维数据的统计分析是现在国际统计学界的热点。非参数经验贝叶斯假定先验分布完全未知,仍然试图逼近最优贝叶斯估计。非参数经验贝叶斯适用于具有相同统计结构的高维未知参数的统计推断问题。美国科学院院士Bradley Efron指出"由于现代数据采集技术和计算机计算能力的快速提高,当今的科学潮流有利于增强非参数经验贝叶斯所起的角色。"本项目主要包括三个子项目:(1)高维信号探测的非参数经验贝叶斯方法,(2)自适应非参数回归中的非参数经验贝叶斯方法,(3)改进以核函数方法构造的经验贝叶斯估计。其中第一个子项目是重点,用非参数极大似然经验贝叶斯方法构造的似然比检验对于高维正态信号不全部为零这一备择假设具有高度的灵敏性,为此需要研究该似然比检验量的理论渐进分布。这个问题同时也具有高度的基础研究价值。预期在三年的研究期限内每年解决一个子问题并发表研究论文。
中文关键词: 高维数据;非参数经验贝叶斯;稀疏正态均值;探测边界;
英文摘要: The purpose of this project is to develop the nonparametric empirical Bayes methods for high-dimensional data. Nonparametric empirical Bayes (Robbins, 1956) assumes essentially no knowledge about the unknown parameters but still aims to attain the performance of the optimal Bayes estimator based on the knowledge of the empirical distribution of the unknowns. Nonparametric empirical Bayes methods are suitable for a sequence of independent statistical decision problems of the same form. Efron (2003) pointed out that "current scientific trends favor a greatly increased role for empirical Bayes methods" due to the prevalence of large, high-dimensional data and rapid rise of computing power. There are three main parts of this project: (1) nonparametric empirical Bayes methods for high-dimensional signal detection; (2) nonparametric empirical Bayes methods for adaptive nonparametric regression; (3) improving kernel empirical Bayes estimation of normal means. The first part is most challenge. It involves deriving the asymptotic distribution of likelihood ration test statistic based on nonparametric maximum likelihood estimator. Preliminary simulation shows that this likelihood ration test is very sensitive to the alternative hypothesis that not all the signals are zero. We expect to solve one part each year.
英文关键词: high-dimensional data;nonparametric empirical Bayes;sparse normal means;detection boundary;