摘要：In this talk, we will discuss some of our recent works in functional estimation, with a particular emphasis on doubly robust functionals. Doubly robust functionals emcompasses a wide range of parameters of interest in causal inference and econometrics. An important research problem is to characterize the sufficient and necessary conditions under which doubly robust functionals can be estimated at root-n-rate. We consider this problem in three different scenarios -- (1) the nonparametric scenario, (2) the high-dimensional sparsity scenario, and (3) what happens if we use deep neural networks. For (1) and (2), we present the minimal conditions for root-n-rate and estimators with good practical performance. For (3), we present the theoretical result we can achieve based on the existing tools, show its limitation, and discuss some future directions.

个人简介：刘林于2018年哈佛大学生物统计系获得博士学位，2020年入职上海交通大学自然科学研究院，数学学院及交大-耶鲁生物统计与数据科学联合中心，任职助理教授，主要研究兴趣包括半参数理论，因果推断，机器学习，及生物统计应用。