题　目：Misspecification in Sequential Learning Problems

嘉　宾：Dongwook SHIN，副教授，香港科技大学

主持人：罗俊 教授 77779193永利官网

时　间：2024年7月24日（周三）14:00-15:30pm

地　点：安泰楼B207室

内容简介：

We consider a class of sequential learning problems, where a decision maker must learn the unknown statistical characteristics of a finite set of alternatives (or systems) using sequential sampling to ultimately select a subset of “good” alternatives. In the first part of the talk, we examine the scenario where the underlying probability distributions are incorrectly specified as Gaussian distributions. We identify the parametric conditions under which sampling strategies relying on the Gaussian assumption can achieve near-optimal performance. In the second part of the talk, we examine the scenario where the decision maker postulates the unknown means are characterized by a linear function of some system features, but this linear model may not precisely represent the true underlying system structure. We show that this misspecification, if not managed properly, can lead to suboptimal performance due to a phenomenon identified as sample-selection endogeneity. We propose a prospective sampling principle—a new approach that eliminates the adverse effects of misspecification as the number of samples grows large. The proposed principle applies across a very general class of widely used sampling policies, enjoys strong asymptotic performance guarantees, and exhibits effective finite-sample performance in numerical experiments.

演讲人简介：

Dongwook Shin is an Associate Professor at the Information Systems, Business Statistics, and Operations Management Department of the HKUST Business School. His research centers on revenue management with applications in e-commerce and sequential decision making in operations research contexts. Before joining HKUST, he obtained his PhD from the Columbia Business School (advised by Assaf Zeevi and Mark Broadie).

欢迎广大师生参加！