报告题目一：Optimal subsampling for high-dimensional partially linear models

 via machine learning methods

报告人：王磊 教授

报告时间：2026年4月24日14点30分

报告地点：文理楼290会议室

报告摘要：In this paper, we explore optimal subsampling strategies for estimating the parametric regression coefficients in partially linear models with unknown nuisance functions involving high-dimensional and potentially endogenous covariates. To address model misspecifications and the curse of dimensionality, we leverage flexible machine learning (ML) techniques to estimate the unknown nuisance functions.  By constructing an unbiased subsampling Neyman-orthogonal score function, we eliminate regularization bias. A two-step algorithm is then used to obtain appropriate ML estimators of the nuisance functions, mitigating the risk of over-fitting. Using martingale techniques, we establish the unconditional consistency and asymptotic normality of the subsample estimators. Furthermore, we derive optimal subsampling probabilities, including A-optimal and L-optimal probabilities as special cases. The proposed optimal subsampling approach is extended to partially linear instrumental variable models to account for potential endogeneity through instrumental variables. Simulation studies and an empirical analysis of the Physicochemical Properties of Protein Tertiary Structure dataset demonstrate the superior performance of our subsample estimators.

报告人简介：王磊，南开大学统计与数据科学学院教授、博导、百名青年学科带头人。研究方向是统计学习和复杂数据分析，已在统计学期刊Biometrika、Jourmal of Machine Learning Research、IEEE Transactions on Information Theory、Annals of Applied Statistics、Bernoulli、Journal of Computational and Graphical Statistics、Statistica Sinica，Science China Mathematics等发表学术论文100多篇，主持3项国家自然科学基金和1项天津市自然科学基金项目。现任中国场统计研究会生存分析分会副秘书长，Journal of Nonparametric Statistics Associate Editor，泛华统计协会永久会员。

报告题目二：Average quantile regression: a new non-mean regression model and coherent risk measure

报告人：姜荣 教授

报告时间：2026年4月24日15点

报告地点：文理楼290会议室

报告摘要：In this talk, I will introduce the innovative concept of Average Quantile Regression (AQR), which is smooth at the quantile-like level, comonotonically additive, and explicitly accounts for the severity of tail losses relative to quantile regression. AQR serves as a versatile regression model capable of describing distributional information across all positions, akin to quantile regression, yet offering enhanced interpretability compared to expectiles. Numerous traditional regression models and coherent risk measures can be regarded as special cases of AQR. As a flexible non-parametric regression model, AQR demonstrates outstanding performance in analyzing high-dimensional and large datasets, particularly those generated by distributed systems, and provides a convenient framework for their statistical analysis. The corresponding estimators are rigorously derived, and their asymptotic properties are thoroughly developed. In a risk management context, the case study confirms AQR’s effectiveness in risk assessment and portfolio optimization.

报告人简介：姜荣，博士，上海对外经贸大学统计与数据科学学院，教授，中国现场统计研究会旅游大数据分会常务理事、全国工业统计学教学研究会金融科技与大数据技术分会理事。主要从事大数据建模、分位数回归和在风险管理中的应用研究。在《Journal of the Royal Statistical Society: Series B》、《Journal of Business & Economic Statistics》、《Journal of the Royal Statistical Society: Series C》、《Journal of Financial Econometrics》、《Test》、《Neurocomputing》和《Journal of Multivariate Analysis》等国际期刊上发表SCI和SSCI论文30余篇。主持国家自然科学基金、教育部人文社科基金和上海市扬帆计划等项目。

报告题目三：Testing for large-dimensional covariance matrix under differential privacy

报告人：朱学虎 教授

报告时间：2026年4月24日15点30分

报告地点：文理楼290会议室

报告摘要：The increasing prevalence of high-dimensional data across various applications has raised significant privacy concerns in statistical inference. In this paper, we propose a differentially private likelihood ratio statistic for testing large-dimensional covariance structures, enabling accurate statistical insights while safeguarding privacy. First, we analyze the global sensitivity of sample eigenvalues for sub-Gaussian populations, where our method bypasses the commonly assumed boundedness of data covariates. For sufficiently large sample size, the privatized statistic guarantees privacy with high probability. Furthermore, when the ratio of dimension to sample size, $d/n \to y \in (0, \infty)$, the privatized statistic is asymptotically normal under the null hypothesis and detects the local alternative hypotheses distinct from the null at the rate of $1/\sqrt{n}$. Extensive numerical studies on synthetic and real data showcase the validity and powerfulness of our proposed method.

报告人简介：朱学虎，博士，西安交通大学教授，博士生导师，主要从事统计学习、高维数据分析及应用统计等领域的研究，截至目前，已发表学术论文30余篇，包括Journal of the American Statistical Association、Journal of Business & Economic Statistics、Science China Mathematics等期刊以及应用领域权威期刊IEEE Transactions on Geoscience and Remote Sensing、计算机顶级会议NeurIPS等。先后主持科技部重点研发计划子课题、国家自然科学基金面上项目、国家自然科学基金重点项目课题等，入选陕西省高校青年杰出人才支持计划和仲英青年学者等荣誉。

报告题目四：Regularized Smoothed Support Tensor Machine

报告人：赵为华 教授

报告时间：2026年4月24日16点

报告地点：文理楼290会议室

报告摘要：Tensor analysis methods are becoming increasingly prevalent across various scientific applications, including neuroscience and signal processing. Existing tensor discrimination models often rely on decomposition techniques such as CP and Tucker decomposition. However, these methods typically require unfolding of tensors into matrices, which may compromise their intrinsic structural information. This article harnesses the recently introduced concept of tubal rank to present a smoothed support tensor machine with tubal nuclear norm regularization.  The statistical properties of the resulting estimator are established, and the framework is extended to a distributed setting. Within this paradigm, a communication-efficient regularized estimator is introduced, which only needs access to local data from the first machine and gradient information from other local machines. Furthermore, the convergence rate of this distributed estimator is derived. By exploiting the well-defined properties of the tubal nuclear norm, we provide theoretical guarantees for low-rank structure recovery. To compute the estimator, an alternating minimization algorithm is developed, and its global convergence properties are analyzed. Lastly, extensive simulations are carried out to validate the proposed method, and its practical utility is demonstrated in an application involving data from invasive ductal carcinoma.

报告人简介：南通大学数学与统计学院教授、博士生导师，江苏省统计科学研究基地主任(南通大学)、统计学科带头人，兼任中国现场统计学会大数据分会常务理事、江苏省概率统计学会常务理事等。主要研究方向：分位数回归、半参数建模、机器学习、张量数据分析。主持完成多项国家级、省部级项目,目前主持国家社科基金一般项目1 项。以第一作者或通讯作者在国内外发表SCI 、SSCI、CSSCI 检索论文 60 多篇。

报告题目五：Double-Robust Small Area Estimation

报告人：马海强 副教授

报告时间：2026年4月24日16点30分

报告地点：文理楼290会议室

报告摘要：In the context of robust small area estimation (SAE), there are two types of robustness considerations, robustness against model misspecification and robustness against outliers. We propose a method of SAE that has both types of robustness features. The method combines the idea of observed best prediction (OBP), which is known to be more robust against model misspecification than the traditional best linear unbiased prediction (EBLUP) method, and the method of density power divergence (DPD), which is known to be more robust against outliers than the EBLUP. The double robust predictor (DRP) is developed under an area-level model with normal or normal-mixture sampling errors, and under a unit-level model. Another advantage of the DRP method is that it provides an natural estimator of a tuning parameter involved in the DPD. We develop theory about the proposed method, and demonstrate empirical performance of the proposed DRP and its comparison to EBLUP, OBP, a robust version of the EBLUP, and predictors based on the DPD. A second order unbiased estimator of the mean squared prediction error of the DRP is developed and its performance is evaluated. A real-data example is discussed. Supplementary materials for this article are available online.

报告人简介：马海强，江西财经大学统计与数据科学学院副教授，博士生导师，主要的研究方向有：函数型数据分析，混合效应模型，稳健统计分析，目前，已在统计学顶级期刊《Journal of the American Statistical Association》（统计四大）、《Statistics Sinica》、《TEST》、《中国科学：数学》、《数学学报》等国内外权威期刊发表学术论文三十多篇，兼任中国现场研究会资源与环境分会理事、全国工业统计研究会理事、全国工业统计研究会青年统计学家协会理事，先后主持国家自然科学基金面上项目、国家自然科学基金青年项目、国家自然科学基金地区项目各 1 项，主持中国博士后面上项目 1 项，江西省自然科学基金重点项目 1 项，江西教育厅科技项目 1 项以及江西高校人文社科项目等省级项目多项，参与科技部重点研发项目，国家社科基金重大项目，国家自然科学基金面上项目多项。

报告题目六：Robust statistical inference for matrix auto-regressive models

报告人：胡雪梅 教授

报告时间：2026年4月26日10点

报告地点：文理楼290会议室

报告摘要：Matrix-valued time series with non-normal distributions are often observed in economics, finance and some other fields. Especially for heavy tailed distributions, ordinary least squares(OLS) estimators are badly influenced by outliers. Moment estimators provide possible alternatives to OLS estimators to compensate the sensitivity of estimates towards outliers and obtain reliable estimates. In this paper we firstly replace an OLS loss (LSL) by a general robust loss (RL) and construct a fire-new p-order robust matrix auto-regression (RMAR(p)) modelling non-normal matrix-valued time series. Secondly, we review an alternating iterative weighted least squares estimation for MAR under LSL(AIWLSE-LSL), creatively propose a fire-new alternating iterative generalized moment estimator procedure under a general RL(AIGME-RL) for RMAR(p), further provide two special AIGME-HL and AIGME-HESL procedures under Huber loss (HL) and Huber exponential squared loss (HESL), and establish its asymptotic properties. Simulations show that the proposed AIGME-HL and AIGME-HESL can efficiently deal with RMAR(1) with multivariate $t_{3}/t_{2}/t_{1}$ distributions, and found that AIGME-HESL performs best, AIGME-HL performs second, AILSE-LSL performs third and LSE-LSL(least squares estimation under LSL) for VAR(1) (1-order vector auto-regression) performs worst. Finally, we apply the relatively optimal AIGME-HESL to analyze and predict China's local government debt risks and Yangtze River Economic Belt's digital finance situations.

报告人简介：胡雪梅，重庆工商大学数学与统计学院教授，成渝地区双城经济圈建设研究院博士生导师，伦敦政治经济学院国家公派访问学者, 香港科技大学数学系访问学者, 重庆经开区经济运行局副局长（挂职）, 中南大学理学博士，中国科学院数学与系统科学研究院控制论国家重点实验室系统科学博士后，第四批重庆市学术技术带头人(统计学)，第五批重庆市高等学校优秀人才支持计划人选，重庆市“统计学”一流专业、首届“统计学”研究生导师团队（2018）、《随机过程》市级一流线下课程负责人。主要研究方向为高维统计、时序分析、多元分析、因子模型、多类分类、统计学习、趋势预测和张量分析等。目前已在IEEE Transaction on Information Theory、Journal of Multivariate Analysis、Expert Systems with Applications、Statistical Papers、North American Journal of Economics and Finance、Journal of Forecasting、Chinese Annals of Mathematics，Series B等国内外学术期刊发表论文50多篇，其中SCI/SSCI收录32篇，主持或完成国家自然科学基金、教育部人文社规划项目等省部级及以上科研项目15项(重大1项、重点3项)，获得重庆市科学技术奖二等奖1项，主编出版学术专著《高维统计模型的估计理论与模型识别》和《高维数据模型的统计学习方法与预测精度评估》，参编英文专著1部。