题目：Ultra-high dimensional semiparametric dynamic high-order spatial autoregressive models
主讲人：范国良
报告时间：2026年5月22日（周五）15：30
地点：综合楼644会议室
报告人简介：范国良，上海海事大学统计学教授，博士生导师，研究方向：非线性关联分析、高维数据分析、充分降维、非参和半参数方法等。主持国家社科基金及国家自科基金共3项，以及教育部人文社科基金、上海市自然科学基金等省部级项目7项；在中国科学(中、英文版)、Statis. Sinica、Statist.Comput、JMVA、CSDA、EJS、JSPI等国内外重要学术刊物上发表学术论文六十余篇。曾获上海市教学成果奖二等奖（排名第一）、安徽省科学技术奖（排名第一）。
讲座摘要：Motivated by the need to effectively characterize complex spatial dependencies inherent in ultra-high dimensional data, this paper develops a sparse semiparametric framework for modeling dynamic high-order spatial autoregressive processes. In this framework, the number of covariates in the linear component grows at a rate much faster than the sample size under a sparsity assumption, whereas the nonparametric component remains of fixed dimension. The varying coefficients are approximated using B-spline basis functions. To address the endogeneity arising from spatial lag terms, two-stage sieve least squares together with instrumental variable methods are employed. We investigate the theoretical properties of the oracle estimator, assuming that the true sparsity structure is known, and establish its convergence rates and asymptotic normality. Further, we propose a nonconvex penalized estimation procedure that simultaneously performs variable selection and estimates both the linear and spatial autoregressive parameters, and we show that it possesses the oracle property under mild conditions. The effectiveness of the proposed method is demonstrated through simulation studies and an empirical application to the Communities and Crime data set from the UCI Machine Learning Repository.
