受国际合作部国际合作与交流资金资助，应002资讯网田波平教授的邀请，美国加州大学河滨分校（University of California, Riverside）统计系马舒洁副教授将于2019年6月20日—2019年6月25日来访公司，并做3场学术讲座和2次座谈，欢迎感兴趣的师生参加。

报告1：Heterogeneity   and Subgroup analysis via non-convex fusion penalization

时间：2019年6月21日9:00-10:30

地点：格物楼503

摘要：Understanding treatment heterogeneity is essential to   the development of precision medicine, which seeks to tailor medical   treatments to subgroups of patients with similar characteristics. One of the   challenges of achieving this goal is that we usually do not have a priori   knowledge of the grouping information of patients with respect to treatment   effect. To address this problem, we consider a heterogeneous regression model   which allows the coefficients for treatment variables or the means to be   subject-dependent with unknown grouping information. We develop a concave   fusion penalized method for estimating the grouping structure and the   subgroup-specific treatment effects. This procedure automatically divides the   observations into subgroups. We develop an alternating direction method of   multipliers algorithm with concave penalties to implement the proposed   approach and demonstrate its convergence. We also establish the theoretical   properties of our proposed estimator and determine the order requirement of   the minimal difference of signals between groups in order to recover them.   These results provide a sound basis for making statistical inference in   subgroup analysis. Moreover, I will talk about applications of our approach   to both cross-sectional data and longitudinal data settings.  This talk   is based on the papers Ma and Huang (2017, JASA) and Ma, Huang, Zhang and Liu   (2019, IJB, revision resubmitted).

座谈1：个人科研经历与培养员工经验浅谈

时间：2019年6月21日10:30-12:00

地点：格物楼503

报告2：A robust and   efficient approach to treatment effect estimation based on sparse sufficient   dimension reduction

时间：2019年6月21日14:30-16:00

地点：明德楼（数学研究院）201室

摘要：A fundamental assumption used in causal inference with   observational data is that treatment assignment is ignorable given measured   confounding variables. This assumption of no missing confounders is plausible   if a large number of baseline covariates are included in the analysis, as we   often have no prior knowledge of which variables can be important   confounders. Thus, estimation of treatment effects with a large number of   covariates has received considerable attention in recent years. Most existing   methods require specifying certain parametric models involving the outcome,   treatment and confounding variables, and employ a variable selection   procedure to identify confounders. However, selection of a proper set of   confounders depends on correct specification of the working models. The bias   due to model misspecification and incorrect selection of confounding   variables can yield misleading results. In this talk, I will talk about a   robust and efficient approach we have proposed for inference about the   average treatment effect via a flexible modeling strategy incorporating   penalized variable selection. Specifically, we consider an estimator   constructed based on an efficient influence function that involves a   propensity score and an outcome regression. We then propose a new sparse   sufficient dimension reduction method to estimate these two functions without   making restrictive parametric modeling assumptions. The proposed estimator of   the average treatment effect is asymptotically normal and semi-parametrically   efficient without the need for variable selection consistency. In the end, I   will talk about simulation studies and a biomedical application. This talk is   based on the paper Ma, Zhu, Zhang, Tsai and Carroll (2018, AoS).

报告3：Estimation and   Inference in Semiparametric Quantile Factor Models

时间：2019年6月22日8:30-10:00

地点：格物楼503

摘要：In this talk, I will introduce an estimation   methodology for a semiparametric quantile factor panel model. I will also   talk about our proposed tools for inference that are robust to the existence   of moments and to the form of weak cross-sectional dependence in the   idiosyncratic error term. Specifically, we use sieve techniques to obtain   preliminary estimators of the nonparametric beta functions, and use these to   estimate the factor return vector at each time period. We then update the   loading functions and factor returns sequentially. We derive the limiting   properties of our estimated factor returns and factor loading functions under   weak conditions on cross-section and temporal dependence. Lastly, I will talk   about applications of our method to daily stock return data. This is a joint   work with Oliver Linton and Jiti Gao.

座谈2：国内外科研环境亲身经历对比浅谈

时间：2019年6月22日10:00-11:30

地点：格物楼503

报告人简介：马舒洁，美国加州大学河滨分校统计系副教授。主要研究精准医学，因子模型，大规模数据分析，高维数据、函数型数据与非线性时间序列数据的统计推断，及其在基因与环境交互作用、环境风险评估、医学与金融数据中的应用等。已在Annals of   Statistics、Journal of the   American Statistical Association、Journal of   Econometrics、Statistics in   Medicine、Bernoulli、Statistica Sinica等期刊上发表论文30余篇，并在国际上已累计做学术会议报告超过30次。现任Journal of Business & Economic Statistics、Computational Statistics and Data Analysis、The American Statistician、Statistica Sinica、Journal of Statistical Planning and Inference期刊的副主编。