报告名称：Variable screening for censored survival data

主办单位：英国立博官网中文版

报告专家：徐锦峰

专家所在单位：香港大学理学院

报告时间：2020年7月29日15:00-16:00

报告地点：zoom(3550318173)

专家简介：徐锦峰，博士，副教授，现为香港大学理学院统计与精算学系老师。主要研究领域为生存分析、生物统计和高维数据。现参与国家自然科学基金（面上项目）1项，主持香港特区政府科研基金项目2项。在统计，生物统计与经济计量学等国内外重要学术期刊上发表科研论文60余篇。

报告摘要：Existing screening procedures for right censored data either posit a specified model or adopt a marginal approach, hence prone to model misspecification or erroneous screening. To address these problems, we develop a joint feature screening method in nonparametric transformation model for censored survival data. A sparsity-restricted estimator is proposed using a smoothed partial rank objective function and an iterative hard thresholding algorithm. We rigorously show that with probability tending to 1, it is capable of retaining all relevant features in the model and more desirable than marginal screening. Furthermore, because the transformation model encompasses many popular models such as the Cox model as special cases, the developed joint screening methodis more robust than its competitors. Its finite sample performance is illustrated using both simulation studies and a real data example.

邀请人：刘展