Statistical Principles and Deep Modeling

 Speaker: 刘传海 DateTime: 2020年12月17日（周四）上午10:30-11:30 Brief Introduction to Speaker: 刘传海，美国普渡大学统计系教授。武汉大学概率统计硕士，1987年、哈佛大学统计学硕士，1990年、哈佛大学统计学博士，1994年。主要研究兴趣包含：贝叶斯、统计推断的计算方法、数据分析的计算机语言和环境、缺失数据和多重插补、多重比较和时间序列等等。获得的奖项/荣誉有美国统计协会会员（2007年）、当选为国际统计学会成员（2006年）、杰出统计应用论文(《美国统计协会杂志》，2000年）、2000年弗兰克·威尔科克森奖、1998年贝尔实验室总裁银奖、1994年哈佛大学杰出教学研究员等等。 Place: 腾讯会议（会议号请联系左国新老师索取） Abstract: While the development of machine learning methods has dominated the recent research in the area of computer-intensive data analysis, one can imagine that future high-quality of research appears to be also in need of more principled ways of data analysis. In this talk, we will introduce the two obvious but fundamental principles, namely, the {\it Validity} principle and the {\it Efficiency} principle. In their book entitled {\it Inferential Models --- Reasoning with Uncertainty}, Ryan Martin and Chuanhai Liu argued for these two principles in the context of making reliable and efficient inference based on postulated models. With a brief review of the two principles for statistical inference, we discuss their implications in a scenario of model building. Implementation of the two principles for model building will be illustrated with an experimental method, which we call {\it Deep Modeling}, for analyzing the famous MNIST dataset.