时 间：2026年4月9日（周四）15：30 - 16：30
地 点：普陀校区理科大楼A1514室
报告人：吴远山 中南财经政法大学教授
主持人：於州 华东师范大学教授
摘 要：
Individualized treatment rules (ITR) typically optimize global expected reward or overall misclassification, which may be inadequate when treatment assignment errors carry asymmetric consequences and lead to degenerate policies. We propose a risk-controlled optimal individualized treatment rule that explicitly distinguishes between two types of treatment mismatches by maximizing policy value for the target subgroup while imposing an explicit risk constraint to protect other subgroups. This constrained learning problem admits a Lagrangian reformulation, where risk control induces asymmetric weighting across treatment groups and acts as a regularization mechanism. We establish theoretical guarantees for the learned treatment rule, including Fisher consistency and an excess risk bound for the policy value of the target subgroup. Extensive simulations and a real example analysis demonstrate that the proposed method improves personalization while maintaining controlled risk compared with existing ITR approaches.
报告人简介：
现任中南财经政法大学统计与数学学院教授、博士生导师。主要从事大数据的统计方法与理论基础、分位数回归、生存分析等相关的研究工作，相关研究成果在Journal of the American Statistical Association、Biometrika、Journal of Machine Learning Research、Science China Mathematics等重要期刊上发表。目前担任ACM Transactions on Probabilistic Machine Learning 的编委。曾多次访问香港大学统计与精算学系等。