Efficient Learning of Optimal Individualized Treatment Rules
时间：2022-09-28
阅读量：574次
- 演讲人： 莫伟斌（普渡大学西拉法叶分校）
- 时间：2022年10月14日 周五上午10:00
- 地点：腾讯会议 ID：255-961-273
- 主办单位：浙江大学数据科学研究中心
摘要：Recent development in data-driven decision science has seen great advancement in individualized decision making. Given the data with individual covariates, treatment assignments and outcomes, researchers can search for the optimal individualized treatment rule (ITR) that maximizes the expected outcome. Existing methods typically require initial estimation of some nuisance models. The double robustness property that can protect consistency from one nuisance model misspecification out of two is widely advocated. However, when model misspecification exists, especially when the outcome model is misspecified, a doubly robust estimate can be sub-optimal. In this presentation, we discuss such an efficiency loss from a heteroscedasticity perspective. To guarantee optimality under this scenario, we propose an Efficient Learning (E-Learning) framework in the multi-categorical treatment setting. We establish the optimality of E-Learning among the class of doubly robust influence function (IF)-based generalized methods-of-moment (GMM) estimates, which can incorporate most regression-based methods in the ITR literature as special cases.
个人简介：Weibin Mo is Assistant Professor at Purdue University Krannert School of Management. His research interests mainly focus on statistical methodologies in machine learning, personalized decision making, causal inference and semiparametric inference, and robust optimization. The major application scenarios of his research are precision medicine and revenue management and pricing.
联系人：崔逸凡（cuiyf@zju.edu.cn）