Statistical Seminar
Globally-Optimal Greedy Active Sequential Estimation
组织者 / Organizer
吴宇楠
报告人 / Speaker
李小鸥
美国明尼苏达大学统计系
时间 / Time
16:00-17:00, May 9, 2025
地点 / Venue
C654, Shuangqing Complex Building A
Abstract
Modern applications such as computerized adaptive testing, sequential rank aggregation, and heterogeneous data source selection increasingly rely on active sequential estimation to enhance parameter inference. This talk explores the design of adaptive experiment selection rules that maximize estimation accuracy while maintaining computational efficiency. Greedy information-based selection strategies, which optimize information gain one step ahead, are widely used due to their flexibility and broad applicability. However, their optimality in the multidimensional setting remains an open question. This talk addresses this gap by establishing rigorous guarantees for multidimensional active sequential estimation within a unified decision-theoretic framework. We prove that maximum likelihood estimators paired with a class of greedy selection rules achieve consistency, asymptotic normality, and optimal risk performance. Additionally, we extend these results to incorporate early stopping mechanisms. Extensive numerical studies on both synthetic and real-world datasets illustrate the advantages of the proposed methods.
