摘要：Selective predictive inference in online settings is crucial for real-time decision-making, where the goal is to make adaptive prediction intervals while effectively managing resource constraints. This talk presents two closely related topics, emphasizing control post-selection coverage and making selection with general constraints in an online manner. In this talk, I will firstly discuss how to construct adaptive pick rule for post-selection conformal prediction with False Coverage-statement Rate (FCR) control, and then introduce a framework for optimizing sample selection through predictive inference while meets both individual and interactive constraints. Theoretical and empirical results validate that these methods offer robust control over coverage rates and selection constraints.

个人简介：

任好洁是上海交通大学数学科学学院长聘教轨副教授，18年博士毕业于南开大学，随后在宾州州立大学从事博士后研究。她的研究方向包括预测推断、统计异常探查、在线学习与监控、高维数据推断等。在JASA，Biometrika等杂志和机器学习顶会ICML，NeurIPS上发表多篇学术论文。