Abstract: This talk presents a unified research program on subgroup learning — statistical methods for extracting maximum information from limited data by exploiting latent structure. We address four problems: (1) Homogeneity Fusion via L₀-constrained Mixed Integer Optimization, which groups regression coefficients into K₀ distinct values and achieves minimax-optimal estimation error scaling with K₀ rather than the sparsity level; (2) Spatial Effect Detection Regression (SEDR), which identifies irregularly bounded active spatial regions in spatiotemporal functional data and exhibits a surprising "curse-to-blessing" phenomenon where more spatial locations improve estimation rates; (3) Center-Augmented Regularization for sample-level subgroup identification, offering a computationally efficient alternative to pairwise penalties and finite mixture models; and (4) a minimax theory comparing subgroup models and threshold regression, revealing a four-phase SNR-dependent behavior and providing principled model selection guidance.

Bio：周岭，西南财经大学光华英才工程特聘教授、博士生导师。四川大学数学学院获得本科和硕士学位，西南财经大学统计学院获得博士学位，随后在美国密歇根大学进行博士后研究。2017年荣获中国数学会钟家庆数学奖，2019年入选国家级和省级青年人才计划。主要从事深度学习理论、非参数理论与方法、分布式数据分析、数据集成、亚组学习等领域研究。迄今在统计学顶刊、机器学习顶刊以及顶会等共发表学术论文40余篇，包括Annals of Statisitcs (AoS)、Journal of the American Statistical Association (JASA)、Journal of the Royal Statistical Society Serires B: Statistical Methodology (JRSSB)、Journal of Machine Learning Research(JMLR)、Jounral of Econometrics (JoE)、Journal of Business and Economic Statistics (JBES)、Biometrics 、The Annals of Applied Statistics (AoAS) 等高水平专业期刊。主持与参与国家项目6项， 现任Statistica Sinica 、The American Statistician等期刊编委(Associate Editor)，国际统计学会(ISI)当选会员(elected member)等。