您当前的位置:首页 >科学研究 > 学术交流

学术报告七十一:A generalized EMS algorithm for model selection with incomplete data

来源:数学与统计学院     作者:     时间:2018/7/13 17:37:41  0次

数学与统计学院学术报告[2018] 071

(高水平大学建设系列报告190)

报告题目: A generalized EMS algorithm for model selection with incomplete data

报告人: 徐平峰  教授(长春工业大学

报告时间:71516:00-17:00

报告地点: 科技楼501                    

报告内容:

The EMS algorithm is a useful method for model selection with missing data. It performs E-step (Expectation step) and MS-step (Model Selection Step) alternately to find the minimum point of the observed generalized information criteria (GIC). However, sometimes it may not be numerically feasible to perform the MS-step, especially for high dimensional settings. In this paper, we seek only a decrease in the observed generalized information criteria in the MS-step. The resulting method is called a generalized EMS (GEMS) algorithm, which includes the EMS algorithm as a special case. We obtain several numerical convergence results of the GEMS algorithm. A useful special case is that all limit points of the EMS algorithm satisfy a necessary condition of the minimum points of the observed GIC under very weak conditions. We apply the GEMS algorithm for Gaussian graphical model selection and variable selection in generalized linear models with missing data and compare with state of the art methods via numerical experiments.

报告人简历:

徐平峰,长春工业大学数学与统计学院副院长, 教授,博士生导师。获首批吉林省高校科研春苗人才称号。2010年毕业于东北师范大学概率论与数理统计专业,师从郭建华教授。曾赴香港浸会大学、美国威斯康星大学、香港恒生管理学院访问交流。任吉林省现场统计研究会常务理事、中国现场统计研究会计算统计分会理事。从事图模型、缺失数据分析、含潜变量的模型等研究工作。发表SCI论文共8篇,均为第一作者或通讯作者,其中一区1篇、二区5篇,主持国家自然科学基金2项,主持吉林省科技厅项目1项。

欢迎感兴趣的师生参加!

分享到: 0

数学与统计学院研究生工作奖励措施

2016年深圳大学数学与统计学院暨暑期夏令营招生指南