报告时间：2026年10月23日（周五）15:30—18:00

报告地点：舜耕校区实验楼505

主办单位：成人网

协办单位：科研处，国势研究院，统计学博士后科研流动站，中加项目管理委员会

一、 工作论文报告

二、特邀专家报告

1.报告一

题目：Testing for Peer Effects in Heteroskedastic Fixed-Effects Panels

摘要：We study testing for peer effects in linear panel models with individual and time fixed effects when the interaction network is unspecified. We propose a quadratic-form test based on the Gram matrix of residualized peer covariates. A key difficulty is that fixed-effects removal and nuisance estimation destroy the centering of a conventional leave-diagonal-out quadratic form under conditional heteroskedasticity. We introduce a leverage adjustment that restores exact conditional centering and combine it with a residual-based variance estimator. The resulting studentized statistic is asymptotically standard normal as both panel dimensions diverge, without imposing any restriction on their relative growth rates. Under additional conditions, we establish consistency against alternatives. The null theory is further extended to covariates that are weakly dependent over time; the same studentization remains valid without an additional long-run variance estimator. Monte Carlo experiments examine the finite-sample size and power of the proposed test across alternative panel dimensions and forms of heteroskedasticity. An empirical application illustrates its use in testing for peer effects when the underlying interaction network is unavailable.

专家介绍：郭旭，现任北京师范大学统计学院教授，博士生导师。目前主要关注高维回归模型中的假设检验问题也对基于机器学习算法的统计推断感兴趣，有多篇文章发表在统计学和计量经济学国际顶尖期刊包括JRSSB, JASA, Biometrika, JOE和JMLR，担任统计学国际知名期刊JMVA副主编和《应用概率统计》杂志第十届编委。现主持国家自然科学基金青年科学基金项目（B类）。曾荣获北京师范大学第十一届“最受本科生欢迎的十佳教师”，北京师范大学第十八届“青教赛”一等奖和北京市第十三届“青教赛”三等奖。

2.报告二

题目：Robust Subsampling for Large-Scale Linear Regression Based on Minimum Distance Kernel Estimation

摘要：Outliers in both the response and covariates make robust subsampling for massive data linear regression especially challenging. Although several robust subsampling approaches have been developed, minimum distance kernel estimation (MDKE), which possesses attractive robustness properties, has not yet been systematically investigated in this context. We fill this gap by developing a computationally efficient subsampling algorithm based on MDKE with product weights, and we establish the consistency and asymptotic normality of the resulting subsample estimators. A repeated-perturbation variant is naturally amenable to parallel computing and attains the full-sample convergence rate. Simulation studies and a real-data analysis show that the proposed method compares favorably with competing robust subsampling methods under heavy-tailed errors and covariate contamination.

专家介绍：王明秋，曲阜师范大学统计与数据科学学院教授，博士生导师。研究方向包括稳健估计、非参数统计推断、高维数据分析、大数据抽样等。主持多项国家自然科学基金与省部级基金项目。主要成果发表在SCIENCE CHINA Mathematics、《统计研究》、Statistics and Computing、Journal of Complexity等国内外重要学术期刊。 指导的研究生多次获山东省优秀硕士学位论文、山东省研究生创新成果奖。

3.报告三

题目：Group-Adversarial Transfer for Causal Effect Estimation: Aligning Domains without Compromising Target-Domain Causal Information

摘要：Estimating average treatment effects (ATEs) is an important task in causal learning. When data in the target domain is scarce, the performance of any causal learning method can be greatly compromised. A remedy of such a challenge is to transfer information from data-rich source domains. However, domain discrepancy can lead to negative transfer, which further harms causal learning performance. Effective adaptation of source data in causal learning requires reducing domain discrepancies, while respecting the treatment-control structure and retaining information relevant to treatment effects. We propose GADANet, a novel transfer causal learning framework that integrates orthogonal representation learning and treatment-control group specific domain alignments. In particular, the orthogonal representation component of GADANet represents the feature by a domain-alignment encoder that seeks the matching of feature distributions between source and target domains, along with a treatment-effect encoder. An orthogonality loss is used to encourage decorrelation between the two encoders. In the domain alignment component of GADANet, group-specific discriminators are designed to conduct adversarial learning with the alignment encoder over the source and target domains. This handles domain discrepancies in covariate distribution, achieves independent alignment within each treatment group and avoids cross-treatment matching. A domain adversarial loss is used to calibrate both the alignment encoder and the discriminators. An outcome prediction loss is further incorporated to capture the causal effect. We evaluated GADANet on six simulated datasets with varying target sizes and shift magnitudes, as well as on the semi-synthetic IHDP benchmark. GADANet achieves significantly better performance compared to the baseline methods. In particular, on the IHDP benchmark, in terms of estimation of average treatment effect, our method reduces the ATE estimation error by 69%–77% relative to the baseline and by 25%–82% relative to Warmstart transfer baseline. These improvements demonstrate that our method is capable of mitigating data scarcity in the target domain and preventing negative transfer under distribution shift. Ablation studies confirm the necessity of each component of our method.

专家介绍：王文武，曲阜师范大学统计与数据科学学院教授、博士生导师;校统计学研究所副所长，院研究生工作负责人。主要从事非参数统计、稳健统计、统计机器学习与因果效应评估等方面研究工作；在机器学习顶级期刊Journal of Machine Learning Research、Knowledge-Based Systems，统计学权威期刊Journal of the American Statistical Association、Statistics in Medicine、Test，心理学权威期刊Psychometrika、Structural Equation Modeling，环境顶级期刊Journal of Hazardous Materials，中文期刊《统计研究》、《数理统计与管理》等发表论文20余篇。目前，主持完成国家自然科学基金面上项目1项、国家统计局重点项目1项；主持教育部人文社科规划项目1项、山东省自然科学基金面上项目1项。曾以高级研究助理、博士后研究员、访问学者、特邀演讲人等身份访问香港大学和香港浸会大学等境外知名高校。